rapidio: add mport char device driver
authorAlexandre Bounine <alexandre.bounine@idt.com>
Tue, 22 Mar 2016 21:27:08 +0000 (14:27 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 22 Mar 2016 22:36:02 +0000 (15:36 -0700)
Add mport character device driver to provide user space interface to
basic RapidIO subsystem operations.

See included Documentation/rapidio/mport_cdev.txt for more details.

[akpm@linux-foundation.org: fix printk warning on i386]
[dan.carpenter@oracle.com: mport_cdev: fix some error codes]
Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Tested-by: Barry Wood <barry.wood@idt.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Aurelien Jacquiot <a-jacquiot@ti.com>
Cc: Andre van Herk <andre.van.herk@prodrive-technologies.com>
Cc: Barry Wood <barry.wood@idt.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Documentation/rapidio/mport_cdev.txt [new file with mode: 0644]
drivers/rapidio/Kconfig
drivers/rapidio/devices/Makefile
drivers/rapidio/devices/rio_mport_cdev.c [new file with mode: 0644]
include/linux/rio_mport_cdev.h [new file with mode: 0644]
include/uapi/linux/Kbuild

diff --git a/Documentation/rapidio/mport_cdev.txt b/Documentation/rapidio/mport_cdev.txt
new file mode 100644 (file)
index 0000000..20c120d
--- /dev/null
@@ -0,0 +1,104 @@
+RapidIO subsystem mport character device driver (rio_mport_cdev.c)
+==================================================================
+
+Version History:
+----------------
+  1.0.0 - Initial driver release.
+
+==================================================================
+
+I. Overview
+
+This device driver is the result of collaboration within the RapidIO.org
+Software Task Group (STG) between Texas Instruments, Freescale,
+Prodrive Technologies, Nokia Networks, BAE and IDT.  Additional input was
+received from other members of RapidIO.org. The objective was to create a
+character mode driver interface which exposes the capabilities of RapidIO
+devices directly to applications, in a manner that allows the numerous and
+varied RapidIO implementations to interoperate.
+
+This driver (MPORT_CDEV) provides access to basic RapidIO subsystem operations
+for user-space applications. Most of RapidIO operations are supported through
+'ioctl' system calls.
+
+When loaded this device driver creates filesystem nodes named rio_mportX in /dev
+directory for each registered RapidIO mport device. 'X' in the node name matches
+to unique port ID assigned to each local mport device.
+
+Using available set of ioctl commands user-space applications can perform
+following RapidIO bus and subsystem operations:
+
+- Reads and writes from/to configuration registers of mport devices
+    (RIO_MPORT_MAINT_READ_LOCAL/RIO_MPORT_MAINT_WRITE_LOCAL)
+- Reads and writes from/to configuration registers of remote RapidIO devices.
+  This operations are defined as RapidIO Maintenance reads/writes in RIO spec.
+    (RIO_MPORT_MAINT_READ_REMOTE/RIO_MPORT_MAINT_WRITE_REMOTE)
+- Set RapidIO Destination ID for mport devices (RIO_MPORT_MAINT_HDID_SET)
+- Set RapidIO Component Tag for mport devices (RIO_MPORT_MAINT_COMPTAG_SET)
+- Query logical index of mport devices (RIO_MPORT_MAINT_PORT_IDX_GET)
+- Query capabilities and RapidIO link configuration of mport devices
+    (RIO_MPORT_GET_PROPERTIES)
+- Enable/Disable reporting of RapidIO doorbell events to user-space applications
+    (RIO_ENABLE_DOORBELL_RANGE/RIO_DISABLE_DOORBELL_RANGE)
+- Enable/Disable reporting of RIO port-write events to user-space applications
+    (RIO_ENABLE_PORTWRITE_RANGE/RIO_DISABLE_PORTWRITE_RANGE)
+- Query/Control type of events reported through this driver: doorbells,
+  port-writes or both (RIO_SET_EVENT_MASK/RIO_GET_EVENT_MASK)
+- Configure/Map mport's outbound requests window(s) for specific size,
+  RapidIO destination ID, hopcount and request type
+    (RIO_MAP_OUTBOUND/RIO_UNMAP_OUTBOUND)
+- Configure/Map mport's inbound requests window(s) for specific size,
+  RapidIO base address and local memory base address
+    (RIO_MAP_INBOUND/RIO_UNMAP_INBOUND)
+- Allocate/Free contiguous DMA coherent memory buffer for DMA data transfers
+  to/from remote RapidIO devices (RIO_ALLOC_DMA/RIO_FREE_DMA)
+- Initiate DMA data transfers to/from remote RapidIO devices (RIO_TRANSFER).
+  Supports blocking, asynchronous and posted (a.k.a 'fire-and-forget') data
+  transfer modes.
+- Check/Wait for completion of asynchronous DMA data transfer
+    (RIO_WAIT_FOR_ASYNC)
+- Manage device objects supported by RapidIO subsystem (RIO_DEV_ADD/RIO_DEV_DEL).
+  This allows implementation of various RapidIO fabric enumeration algorithms
+  as user-space applications while using remaining functionality provided by
+  kernel RapidIO subsystem.
+
+II. Hardware Compatibility
+
+This device driver uses standard interfaces defined by kernel RapidIO subsystem
+and therefore it can be used with any mport device driver registered by RapidIO
+subsystem with limitations set by available mport implementation.
+
+At this moment the most common limitation is availability of RapidIO-specific
+DMA engine framework for specific mport device. Users should verify available
+functionality of their platform when planning to use this driver:
+
+- IDT Tsi721 PCIe-to-RapidIO bridge device and its mport device driver are fully
+  compatible with this driver.
+- Freescale SoCs 'fsl_rio' mport driver does not have implementation for RapidIO
+  specific DMA engine support and therefore DMA data transfers mport_cdev driver
+  are not available.
+
+III. Module parameters
+
+- 'dbg_level' - This parameter allows to control amount of debug information
+        generated by this device driver. This parameter is formed by set of
+        This parameter can be changed bit masks that correspond to the specific
+        functional block.
+        For mask definitions see 'drivers/rapidio/devices/rio_mport_cdev.c'
+        This parameter can be changed dynamically.
+        Use CONFIG_RAPIDIO_DEBUG=y to enable debug output at the top level.
+
+IV. Known problems
+
+  None.
+
+V. User-space Applications and API
+
+API library and applications that use this device driver are available from
+RapidIO.org.
+
+VI. TODO List
+
+- Add support for sending/receiving "raw" RapidIO messaging packets.
+- Add memory mapped DMA data transfers as an option when RapidIO-specific DMA
+  is not available.
index 3e3be57..b5a10d3 100644 (file)
@@ -67,6 +67,14 @@ config RAPIDIO_ENUM_BASIC
 
 endchoice
 
+config RAPIDIO_MPORT_CDEV
+       tristate "RapidIO /dev mport device driver"
+       depends on RAPIDIO
+       help
+         This option includes generic RapidIO mport device driver which
+         allows to user space applications to perform RapidIO-specific
+         operations through selected RapidIO mport.
+
 menu "RapidIO Switch drivers"
        depends on RAPIDIO
 
index 9432c49..927dbf8 100644 (file)
@@ -5,3 +5,4 @@
 obj-$(CONFIG_RAPIDIO_TSI721)   += tsi721_mport.o
 tsi721_mport-y                 := tsi721.o
 tsi721_mport-$(CONFIG_RAPIDIO_DMA_ENGINE) += tsi721_dma.o
+obj-$(CONFIG_RAPIDIO_MPORT_CDEV) += rio_mport_cdev.o
diff --git a/drivers/rapidio/devices/rio_mport_cdev.c b/drivers/rapidio/devices/rio_mport_cdev.c
new file mode 100644 (file)
index 0000000..9607bc8
--- /dev/null
@@ -0,0 +1,2720 @@
+/*
+ * RapidIO mport character device
+ *
+ * Copyright 2014-2015 Integrated Device Technology, Inc.
+ *    Alexandre Bounine <alexandre.bounine@idt.com>
+ * Copyright 2014-2015 Prodrive Technologies
+ *    Andre van Herk <andre.van.herk@prodrive-technologies.com>
+ *    Jerry Jacobs <jerry.jacobs@prodrive-technologies.com>
+ * Copyright (C) 2014 Texas Instruments Incorporated
+ *    Aurelien Jacquiot <a-jacquiot@ti.com>
+ *
+ * This program is free software; you can redistribute  it and/or modify it
+ * under  the terms of  the GNU General  Public License as published by the
+ * Free Software Foundation;  either version 2 of the  License, or (at your
+ * option) any later version.
+ */
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/cdev.h>
+#include <linux/ioctl.h>
+#include <linux/uaccess.h>
+#include <linux/list.h>
+#include <linux/fs.h>
+#include <linux/err.h>
+#include <linux/net.h>
+#include <linux/poll.h>
+#include <linux/spinlock.h>
+#include <linux/sched.h>
+#include <linux/kfifo.h>
+
+#include <linux/mm.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/mman.h>
+
+#include <linux/dma-mapping.h>
+#ifdef CONFIG_RAPIDIO_DMA_ENGINE
+#include <linux/dmaengine.h>
+#endif
+
+#include <linux/rio.h>
+#include <linux/rio_ids.h>
+#include <linux/rio_drv.h>
+#include <linux/rio_mport_cdev.h>
+
+#include "../rio.h"
+
+#define DRV_NAME       "rio_mport"
+#define DRV_PREFIX     DRV_NAME ": "
+#define DEV_NAME       "rio_mport"
+#define DRV_VERSION     "1.0.0"
+
+/* Debug output filtering masks */
+enum {
+       DBG_NONE        = 0,
+       DBG_INIT        = BIT(0), /* driver init */
+       DBG_EXIT        = BIT(1), /* driver exit */
+       DBG_MPORT       = BIT(2), /* mport add/remove */
+       DBG_RDEV        = BIT(3), /* RapidIO device add/remove */
+       DBG_DMA         = BIT(4), /* DMA transfer messages */
+       DBG_MMAP        = BIT(5), /* mapping messages */
+       DBG_IBW         = BIT(6), /* inbound window */
+       DBG_EVENT       = BIT(7), /* event handling messages */
+       DBG_OBW         = BIT(8), /* outbound window messages */
+       DBG_DBELL       = BIT(9), /* doorbell messages */
+       DBG_ALL         = ~0,
+};
+
+#ifdef DEBUG
+#define rmcd_debug(level, fmt, arg...)         \
+       do {                                    \
+               if (DBG_##level & dbg_level)    \
+                       pr_debug(DRV_PREFIX "%s: " fmt "\n", __func__, ##arg); \
+       } while (0)
+#else
+#define rmcd_debug(level, fmt, arg...) \
+               no_printk(KERN_DEBUG pr_fmt(DRV_PREFIX fmt "\n"), ##arg)
+#endif
+
+#define rmcd_warn(fmt, arg...) \
+       pr_warn(DRV_PREFIX "%s WARNING " fmt "\n", __func__, ##arg)
+
+#define rmcd_error(fmt, arg...) \
+       pr_err(DRV_PREFIX "%s ERROR " fmt "\n", __func__, ##arg)
+
+MODULE_AUTHOR("Jerry Jacobs <jerry.jacobs@prodrive-technologies.com>");
+MODULE_AUTHOR("Aurelien Jacquiot <a-jacquiot@ti.com>");
+MODULE_AUTHOR("Alexandre Bounine <alexandre.bounine@idt.com>");
+MODULE_AUTHOR("Andre van Herk <andre.van.herk@prodrive-technologies.com>");
+MODULE_DESCRIPTION("RapidIO mport character device driver");
+MODULE_LICENSE("GPL");
+MODULE_VERSION(DRV_VERSION);
+
+static int dma_timeout = 3000; /* DMA transfer timeout in msec */
+module_param(dma_timeout, int, S_IRUGO);
+MODULE_PARM_DESC(dma_timeout, "DMA Transfer Timeout in msec (default: 3000)");
+
+#ifdef DEBUG
+static u32 dbg_level = DBG_NONE;
+module_param(dbg_level, uint, S_IWUSR | S_IWGRP | S_IRUGO);
+MODULE_PARM_DESC(dbg_level, "Debugging output level (default 0 = none)");
+#endif
+
+/*
+ * An internal DMA coherent buffer
+ */
+struct mport_dma_buf {
+       void            *ib_base;
+       dma_addr_t      ib_phys;
+       u32             ib_size;
+       u64             ib_rio_base;
+       bool            ib_map;
+       struct file     *filp;
+};
+
+/*
+ * Internal memory mapping structure
+ */
+enum rio_mport_map_dir {
+       MAP_INBOUND,
+       MAP_OUTBOUND,
+       MAP_DMA,
+};
+
+struct rio_mport_mapping {
+       struct list_head node;
+       struct mport_dev *md;
+       enum rio_mport_map_dir dir;
+       u32 rioid;
+       u64 rio_addr;
+       dma_addr_t phys_addr; /* for mmap */
+       void *virt_addr; /* kernel address, for dma_free_coherent */
+       u64 size;
+       struct kref ref; /* refcount of vmas sharing the mapping */
+       struct file *filp;
+};
+
+struct rio_mport_dma_map {
+       int valid;
+       uint64_t length;
+       void *vaddr;
+       dma_addr_t paddr;
+};
+
+#define MPORT_MAX_DMA_BUFS     16
+#define MPORT_EVENT_DEPTH      10
+
+/*
+ * mport_dev  driver-specific structure that represents mport device
+ * @active    mport device status flag
+ * @node      list node to maintain list of registered mports
+ * @cdev      character device
+ * @dev       associated device object
+ * @mport     associated subsystem's master port device object
+ * @buf_mutex lock for buffer handling
+ * @file_mutex - lock for open files list
+ * @file_list  - list of open files on given mport
+ * @properties properties of this mport
+ * @portwrites queue of inbound portwrites
+ * @pw_lock    lock for port write queue
+ * @mappings   queue for memory mappings
+ * @dma_chan   DMA channels associated with this device
+ * @dma_ref:
+ * @comp:
+ */
+struct mport_dev {
+       atomic_t                active;
+       struct list_head        node;
+       struct cdev             cdev;
+       struct device           dev;
+       struct rio_mport        *mport;
+       struct mutex            buf_mutex;
+       struct mutex            file_mutex;
+       struct list_head        file_list;
+       struct rio_mport_properties     properties;
+       struct list_head                doorbells;
+       spinlock_t                      db_lock;
+       struct list_head                portwrites;
+       spinlock_t                      pw_lock;
+       struct list_head        mappings;
+#ifdef CONFIG_RAPIDIO_DMA_ENGINE
+       struct dma_chan *dma_chan;
+       struct kref     dma_ref;
+       struct completion comp;
+#endif
+};
+
+/*
+ * mport_cdev_priv - data structure specific to individual file object
+ *                   associated with an open device
+ * @md    master port character device object
+ * @async_queue - asynchronous notification queue
+ * @list - file objects tracking list
+ * @db_filters    inbound doorbell filters for this descriptor
+ * @pw_filters    portwrite filters for this descriptor
+ * @event_fifo    event fifo for this descriptor
+ * @event_rx_wait wait queue for this descriptor
+ * @fifo_lock     lock for event_fifo
+ * @event_mask    event mask for this descriptor
+ * @dmach DMA engine channel allocated for specific file object
+ */
+struct mport_cdev_priv {
+       struct mport_dev        *md;
+       struct fasync_struct    *async_queue;
+       struct list_head        list;
+       struct list_head        db_filters;
+       struct list_head        pw_filters;
+       struct kfifo            event_fifo;
+       wait_queue_head_t       event_rx_wait;
+       spinlock_t              fifo_lock;
+       unsigned int            event_mask; /* RIO_DOORBELL, RIO_PORTWRITE */
+#ifdef CONFIG_RAPIDIO_DMA_ENGINE
+       struct dma_chan         *dmach;
+       struct list_head        async_list;
+       struct list_head        pend_list;
+       spinlock_t              req_lock;
+       struct mutex            dma_lock;
+       struct kref             dma_ref;
+       struct completion       comp;
+#endif
+};
+
+/*
+ * rio_mport_pw_filter - structure to describe a portwrite filter
+ * md_node   node in mport device's list
+ * priv_node node in private file object's list
+ * priv      reference to private data
+ * filter    actual portwrite filter
+ */
+struct rio_mport_pw_filter {
+       struct list_head md_node;
+       struct list_head priv_node;
+       struct mport_cdev_priv *priv;
+       struct rio_pw_filter filter;
+};
+
+/*
+ * rio_mport_db_filter - structure to describe a doorbell filter
+ * @data_node reference to device node
+ * @priv_node node in private data
+ * @priv      reference to private data
+ * @filter    actual doorbell filter
+ */
+struct rio_mport_db_filter {
+       struct list_head data_node;
+       struct list_head priv_node;
+       struct mport_cdev_priv *priv;
+       struct rio_doorbell_filter filter;
+};
+
+static LIST_HEAD(mport_devs);
+static DEFINE_MUTEX(mport_devs_lock);
+
+#if (0) /* used by commented out portion of poll function : FIXME */
+static DECLARE_WAIT_QUEUE_HEAD(mport_cdev_wait);
+#endif
+
+static struct class *dev_class;
+static dev_t dev_number;
+
+static struct workqueue_struct *dma_wq;
+
+static void mport_release_mapping(struct kref *ref);
+
+static int rio_mport_maint_rd(struct mport_cdev_priv *priv, void __user *arg,
+                             int local)
+{
+       struct rio_mport *mport = priv->md->mport;
+       struct rio_mport_maint_io maint_io;
+       u32 *buffer;
+       u32 offset;
+       size_t length;
+       int ret, i;
+
+       if (unlikely(copy_from_user(&maint_io, arg, sizeof(maint_io))))
+               return -EFAULT;
+
+       if ((maint_io.offset % 4) ||
+           (maint_io.length == 0) || (maint_io.length % 4))
+               return -EINVAL;
+
+       buffer = vmalloc(maint_io.length);
+       if (buffer == NULL)
+               return -ENOMEM;
+       length = maint_io.length/sizeof(u32);
+       offset = maint_io.offset;
+
+       for (i = 0; i < length; i++) {
+               if (local)
+                       ret = __rio_local_read_config_32(mport,
+                               offset, &buffer[i]);
+               else
+                       ret = rio_mport_read_config_32(mport, maint_io.rioid,
+                               maint_io.hopcount, offset, &buffer[i]);
+               if (ret)
+                       goto out;
+
+               offset += 4;
+       }
+
+       if (unlikely(copy_to_user(maint_io.buffer, buffer, maint_io.length)))
+               ret = -EFAULT;
+out:
+       vfree(buffer);
+       return ret;
+}
+
+static int rio_mport_maint_wr(struct mport_cdev_priv *priv, void __user *arg,
+                             int local)
+{
+       struct rio_mport *mport = priv->md->mport;
+       struct rio_mport_maint_io maint_io;
+       u32 *buffer;
+       u32 offset;
+       size_t length;
+       int ret = -EINVAL, i;
+
+       if (unlikely(copy_from_user(&maint_io, arg, sizeof(maint_io))))
+               return -EFAULT;
+
+       if ((maint_io.offset % 4) ||
+           (maint_io.length == 0) || (maint_io.length % 4))
+               return -EINVAL;
+
+       buffer = vmalloc(maint_io.length);
+       if (buffer == NULL)
+               return -ENOMEM;
+       length = maint_io.length;
+
+       if (unlikely(copy_from_user(buffer, maint_io.buffer, length))) {
+               ret = -EFAULT;
+               goto out;
+       }
+
+       offset = maint_io.offset;
+       length /= sizeof(u32);
+
+       for (i = 0; i < length; i++) {
+               if (local)
+                       ret = __rio_local_write_config_32(mport,
+                                                         offset, buffer[i]);
+               else
+                       ret = rio_mport_write_config_32(mport, maint_io.rioid,
+                                                       maint_io.hopcount,
+                                                       offset, buffer[i]);
+               if (ret)
+                       goto out;
+
+               offset += 4;
+       }
+
+out:
+       vfree(buffer);
+       return ret;
+}
+
+
+/*
+ * Inbound/outbound memory mapping functions
+ */
+static int
+rio_mport_create_outbound_mapping(struct mport_dev *md, struct file *filp,
+                                 u32 rioid, u64 raddr, u32 size,
+                                 dma_addr_t *paddr)
+{
+       struct rio_mport *mport = md->mport;
+       struct rio_mport_mapping *map;
+       int ret;
+
+       rmcd_debug(OBW, "did=%d ra=0x%llx sz=0x%x", rioid, raddr, size);
+
+       map = kzalloc(sizeof(struct rio_mport_mapping), GFP_KERNEL);
+       if (map == NULL)
+               return -ENOMEM;
+
+       ret = rio_map_outb_region(mport, rioid, raddr, size, 0, paddr);
+       if (ret < 0)
+               goto err_map_outb;
+
+       map->dir = MAP_OUTBOUND;
+       map->rioid = rioid;
+       map->rio_addr = raddr;
+       map->size = size;
+       map->phys_addr = *paddr;
+       map->filp = filp;
+       map->md = md;
+       kref_init(&map->ref);
+       list_add_tail(&map->node, &md->mappings);
+       return 0;
+err_map_outb:
+       kfree(map);
+       return ret;
+}
+
+static int
+rio_mport_get_outbound_mapping(struct mport_dev *md, struct file *filp,
+                              u32 rioid, u64 raddr, u32 size,
+                              dma_addr_t *paddr)
+{
+       struct rio_mport_mapping *map;
+       int err = -ENOMEM;
+
+       mutex_lock(&md->buf_mutex);
+       list_for_each_entry(map, &md->mappings, node) {
+               if (map->dir != MAP_OUTBOUND)
+                       continue;
+               if (rioid == map->rioid &&
+                   raddr == map->rio_addr && size == map->size) {
+                       *paddr = map->phys_addr;
+                       err = 0;
+                       break;
+               } else if (rioid == map->rioid &&
+                          raddr < (map->rio_addr + map->size - 1) &&
+                          (raddr + size) > map->rio_addr) {
+                       err = -EBUSY;
+                       break;
+               }
+       }
+
+       /* If not found, create new */
+       if (err == -ENOMEM)
+               err = rio_mport_create_outbound_mapping(md, filp, rioid, raddr,
+                                               size, paddr);
+       mutex_unlock(&md->buf_mutex);
+       return err;
+}
+
+static int rio_mport_obw_map(struct file *filp, void __user *arg)
+{
+       struct mport_cdev_priv *priv = filp->private_data;
+       struct mport_dev *data = priv->md;
+       struct rio_mmap map;
+       dma_addr_t paddr;
+       int ret;
+
+       if (unlikely(copy_from_user(&map, arg, sizeof(struct rio_mmap))))
+               return -EFAULT;
+
+       rmcd_debug(OBW, "did=%d ra=0x%llx sz=0x%llx",
+                  map.rioid, map.rio_addr, map.length);
+
+       ret = rio_mport_get_outbound_mapping(data, filp, map.rioid,
+                                            map.rio_addr, map.length, &paddr);
+       if (ret < 0) {
+               rmcd_error("Failed to set OBW err= %d", ret);
+               return ret;
+       }
+
+       map.handle = paddr;
+
+       if (unlikely(copy_to_user(arg, &map, sizeof(struct rio_mmap))))
+               return -EFAULT;
+       return 0;
+}
+
+/*
+ * rio_mport_obw_free() - unmap an OutBound Window from RapidIO address space
+ *
+ * @priv: driver private data
+ * @arg:  buffer handle returned by allocation routine
+ */
+static int rio_mport_obw_free(struct file *filp, void __user *arg)
+{
+       struct mport_cdev_priv *priv = filp->private_data;
+       struct mport_dev *md = priv->md;
+       u64 handle;
+       struct rio_mport_mapping *map, *_map;
+
+       if (!md->mport->ops->unmap_outb)
+               return -EPROTONOSUPPORT;
+
+       if (copy_from_user(&handle, arg, sizeof(u64)))
+               return -EFAULT;
+
+       rmcd_debug(OBW, "h=0x%llx", handle);
+
+       mutex_lock(&md->buf_mutex);
+       list_for_each_entry_safe(map, _map, &md->mappings, node) {
+               if (map->dir == MAP_OUTBOUND && map->phys_addr == handle) {
+                       if (map->filp == filp) {
+                               rmcd_debug(OBW, "kref_put h=0x%llx", handle);
+                               map->filp = NULL;
+                               kref_put(&map->ref, mport_release_mapping);
+                       }
+                       break;
+               }
+       }
+       mutex_unlock(&md->buf_mutex);
+
+       return 0;
+}
+
+/*
+ * maint_hdid_set() - Set the host Device ID
+ * @priv: driver private data
+ * @arg:       Device Id
+ */
+static int maint_hdid_set(struct mport_cdev_priv *priv, void __user *arg)
+{
+       struct mport_dev *md = priv->md;
+       uint16_t hdid;
+
+       if (copy_from_user(&hdid, arg, sizeof(uint16_t)))
+               return -EFAULT;
+
+       md->mport->host_deviceid = hdid;
+       md->properties.hdid = hdid;
+       rio_local_set_device_id(md->mport, hdid);
+
+       rmcd_debug(MPORT, "Set host device Id to %d", hdid);
+
+       return 0;
+}
+
+/*
+ * maint_comptag_set() - Set the host Component Tag
+ * @priv: driver private data
+ * @arg:       Component Tag
+ */
+static int maint_comptag_set(struct mport_cdev_priv *priv, void __user *arg)
+{
+       struct mport_dev *md = priv->md;
+       uint32_t comptag;
+
+       if (copy_from_user(&comptag, arg, sizeof(uint32_t)))
+               return -EFAULT;
+
+       rio_local_write_config_32(md->mport, RIO_COMPONENT_TAG_CSR, comptag);
+
+       rmcd_debug(MPORT, "Set host Component Tag to %d", comptag);
+
+       return 0;
+}
+
+#ifdef CONFIG_RAPIDIO_DMA_ENGINE
+
+struct mport_dma_req {
+       struct list_head node;
+       struct file *filp;
+       struct mport_cdev_priv *priv;
+       enum rio_transfer_sync sync;
+       struct sg_table sgt;
+       struct page **page_list;
+       unsigned int nr_pages;
+       struct rio_mport_mapping *map;
+       struct dma_chan *dmach;
+       enum dma_data_direction dir;
+       dma_cookie_t cookie;
+       enum dma_status status;
+       struct completion req_comp;
+};
+
+struct mport_faf_work {
+       struct work_struct work;
+       struct mport_dma_req *req;
+};
+
+static void mport_release_def_dma(struct kref *dma_ref)
+{
+       struct mport_dev *md =
+                       container_of(dma_ref, struct mport_dev, dma_ref);
+
+       rmcd_debug(EXIT, "DMA_%d", md->dma_chan->chan_id);
+       rio_release_dma(md->dma_chan);
+       md->dma_chan = NULL;
+}
+
+static void mport_release_dma(struct kref *dma_ref)
+{
+       struct mport_cdev_priv *priv =
+                       container_of(dma_ref, struct mport_cdev_priv, dma_ref);
+
+       rmcd_debug(EXIT, "DMA_%d", priv->dmach->chan_id);
+       complete(&priv->comp);
+}
+
+static void dma_req_free(struct mport_dma_req *req)
+{
+       struct mport_cdev_priv *priv = req->priv;
+       unsigned int i;
+
+       dma_unmap_sg(req->dmach->device->dev,
+                    req->sgt.sgl, req->sgt.nents, req->dir);
+       sg_free_table(&req->sgt);
+       if (req->page_list) {
+               for (i = 0; i < req->nr_pages; i++)
+                       put_page(req->page_list[i]);
+               kfree(req->page_list);
+       }
+
+       if (req->map) {
+               mutex_lock(&req->map->md->buf_mutex);
+               kref_put(&req->map->ref, mport_release_mapping);
+               mutex_unlock(&req->map->md->buf_mutex);
+       }
+
+       kref_put(&priv->dma_ref, mport_release_dma);
+
+       kfree(req);
+}
+
+static void dma_xfer_callback(void *param)
+{
+       struct mport_dma_req *req = (struct mport_dma_req *)param;
+       struct mport_cdev_priv *priv = req->priv;
+
+       req->status = dma_async_is_tx_complete(priv->dmach, req->cookie,
+                                              NULL, NULL);
+       complete(&req->req_comp);
+}
+
+static void dma_faf_cleanup(struct work_struct *_work)
+{
+       struct mport_faf_work *work = container_of(_work,
+                                               struct mport_faf_work, work);
+       struct mport_dma_req *req = work->req;
+
+       dma_req_free(req);
+       kfree(work);
+}
+
+static void dma_faf_callback(void *param)
+{
+       struct mport_dma_req *req = (struct mport_dma_req *)param;
+       struct mport_faf_work *work;
+
+       work = kmalloc(sizeof(*work), GFP_ATOMIC);
+       if (!work)
+               return;
+
+       INIT_WORK(&work->work, dma_faf_cleanup);
+       work->req = req;
+       queue_work(dma_wq, &work->work);
+}
+
+/*
+ * prep_dma_xfer() - Configure and send request to DMAengine to prepare DMA
+ *                   transfer object.
+ * Returns pointer to DMA transaction descriptor allocated by DMA driver on
+ * success or ERR_PTR (and/or NULL) if failed. Caller must check returned
+ * non-NULL pointer using IS_ERR macro.
+ */
+static struct dma_async_tx_descriptor
+*prep_dma_xfer(struct dma_chan *chan, struct rio_transfer_io *transfer,
+       struct sg_table *sgt, int nents, enum dma_transfer_direction dir,
+       enum dma_ctrl_flags flags)
+{
+       struct rio_dma_data tx_data;
+
+       tx_data.sg = sgt->sgl;
+       tx_data.sg_len = nents;
+       tx_data.rio_addr_u = 0;
+       tx_data.rio_addr = transfer->rio_addr;
+       if (dir == DMA_MEM_TO_DEV) {
+               switch (transfer->method) {
+               case RIO_EXCHANGE_NWRITE:
+                       tx_data.wr_type = RDW_ALL_NWRITE;
+                       break;
+               case RIO_EXCHANGE_NWRITE_R_ALL:
+                       tx_data.wr_type = RDW_ALL_NWRITE_R;
+                       break;
+               case RIO_EXCHANGE_NWRITE_R:
+                       tx_data.wr_type = RDW_LAST_NWRITE_R;
+                       break;
+               case RIO_EXCHANGE_DEFAULT:
+                       tx_data.wr_type = RDW_DEFAULT;
+                       break;
+               default:
+                       return ERR_PTR(-EINVAL);
+               }
+       }
+
+       return rio_dma_prep_xfer(chan, transfer->rioid, &tx_data, dir, flags);
+}
+
+/* Request DMA channel associated with this mport device.
+ * Try to request DMA channel for every new process that opened given
+ * mport. If a new DMA channel is not available use default channel
+ * which is the first DMA channel opened on mport device.
+ */
+static int get_dma_channel(struct mport_cdev_priv *priv)
+{
+       mutex_lock(&priv->dma_lock);
+       if (!priv->dmach) {
+               priv->dmach = rio_request_mport_dma(priv->md->mport);
+               if (!priv->dmach) {
+                       /* Use default DMA channel if available */
+                       if (priv->md->dma_chan) {
+                               priv->dmach = priv->md->dma_chan;
+                               kref_get(&priv->md->dma_ref);
+                       } else {
+                               rmcd_error("Failed to get DMA channel");
+                               mutex_unlock(&priv->dma_lock);
+                               return -ENODEV;
+                       }
+               } else if (!priv->md->dma_chan) {
+                       /* Register default DMA channel if we do not have one */
+                       priv->md->dma_chan = priv->dmach;
+                       kref_init(&priv->md->dma_ref);
+                       rmcd_debug(DMA, "Register DMA_chan %d as default",
+                                  priv->dmach->chan_id);
+               }
+
+               kref_init(&priv->dma_ref);
+               init_completion(&priv->comp);
+       }
+
+       kref_get(&priv->dma_ref);
+       mutex_unlock(&priv->dma_lock);
+       return 0;
+}
+
+static void put_dma_channel(struct mport_cdev_priv *priv)
+{
+       kref_put(&priv->dma_ref, mport_release_dma);
+}
+
+/*
+ * DMA transfer functions
+ */
+static int do_dma_request(struct mport_dma_req *req,
+                         struct rio_transfer_io *xfer,
+                         enum rio_transfer_sync sync, int nents)
+{
+       struct mport_cdev_priv *priv;
+       struct sg_table *sgt;
+       struct dma_chan *chan;
+       struct dma_async_tx_descriptor *tx;
+       dma_cookie_t cookie;
+       unsigned long tmo = msecs_to_jiffies(dma_timeout);
+       enum dma_transfer_direction dir;
+       long wret;
+       int ret = 0;
+
+       priv = req->priv;
+       sgt = &req->sgt;
+
+       chan = priv->dmach;
+       dir = (req->dir == DMA_FROM_DEVICE) ? DMA_DEV_TO_MEM : DMA_MEM_TO_DEV;
+
+       rmcd_debug(DMA, "%s(%d) uses %s for DMA_%s",
+                  current->comm, task_pid_nr(current),
+                  dev_name(&chan->dev->device),
+                  (dir == DMA_DEV_TO_MEM)?"READ":"WRITE");
+
+       /* Initialize DMA transaction request */
+       tx = prep_dma_xfer(chan, xfer, sgt, nents, dir,
+                          DMA_CTRL_ACK | DMA_PREP_INTERRUPT);
+
+       if (!tx) {
+               rmcd_debug(DMA, "prep error for %s A:0x%llx L:0x%llx",
+                       (dir == DMA_DEV_TO_MEM)?"READ":"WRITE",
+                       xfer->rio_addr, xfer->length);
+               ret = -EIO;
+               goto err_out;
+       } else if (IS_ERR(tx)) {
+               ret = PTR_ERR(tx);
+               rmcd_debug(DMA, "prep error %d for %s A:0x%llx L:0x%llx", ret,
+                       (dir == DMA_DEV_TO_MEM)?"READ":"WRITE",
+                       xfer->rio_addr, xfer->length);
+               goto err_out;
+       }
+
+       if (sync == RIO_TRANSFER_FAF)
+               tx->callback = dma_faf_callback;
+       else
+               tx->callback = dma_xfer_callback;
+       tx->callback_param = req;
+
+       req->dmach = chan;
+       req->sync = sync;
+       req->status = DMA_IN_PROGRESS;
+       init_completion(&req->req_comp);
+
+       cookie = dmaengine_submit(tx);
+       req->cookie = cookie;
+
+       rmcd_debug(DMA, "pid=%d DMA_%s tx_cookie = %d", task_pid_nr(current),
+                  (dir == DMA_DEV_TO_MEM)?"READ":"WRITE", cookie);
+
+       if (dma_submit_error(cookie)) {
+               rmcd_error("submit err=%d (addr:0x%llx len:0x%llx)",
+                          cookie, xfer->rio_addr, xfer->length);
+               ret = -EIO;
+               goto err_out;
+       }
+
+       dma_async_issue_pending(chan);
+
+       if (sync == RIO_TRANSFER_ASYNC) {
+               spin_lock(&priv->req_lock);
+               list_add_tail(&req->node, &priv->async_list);
+               spin_unlock(&priv->req_lock);
+               return cookie;
+       } else if (sync == RIO_TRANSFER_FAF)
+               return 0;
+
+       wret = wait_for_completion_interruptible_timeout(&req->req_comp, tmo);
+
+       if (wret == 0) {
+               /* Timeout on wait occurred */
+               rmcd_error("%s(%d) timed out waiting for DMA_%s %d",
+                      current->comm, task_pid_nr(current),
+                      (dir == DMA_DEV_TO_MEM)?"READ":"WRITE", cookie);
+               return -ETIMEDOUT;
+       } else if (wret == -ERESTARTSYS) {
+               /* Wait_for_completion was interrupted by a signal but DMA may
+                * be in progress
+                */
+               rmcd_error("%s(%d) wait for DMA_%s %d was interrupted",
+                       current->comm, task_pid_nr(current),
+                       (dir == DMA_DEV_TO_MEM)?"READ":"WRITE", cookie);
+               return -EINTR;
+       }
+
+       if (req->status != DMA_COMPLETE) {
+               /* DMA transaction completion was signaled with error */
+               rmcd_error("%s(%d) DMA_%s %d completed with status %d (ret=%d)",
+                       current->comm, task_pid_nr(current),
+                       (dir == DMA_DEV_TO_MEM)?"READ":"WRITE",
+                       cookie, req->status, ret);
+               ret = -EIO;
+       }
+
+err_out:
+       return ret;
+}
+
+/*
+ * rio_dma_transfer() - Perform RapidIO DMA data transfer to/from
+ *                      the remote RapidIO device
+ * @filp: file pointer associated with the call
+ * @transfer_mode: DMA transfer mode
+ * @sync: synchronization mode
+ * @dir: DMA transfer direction (DMA_MEM_TO_DEV = write OR
+ *                               DMA_DEV_TO_MEM = read)
+ * @xfer: data transfer descriptor structure
+ */
+static int
+rio_dma_transfer(struct file *filp, uint32_t transfer_mode,
+                enum rio_transfer_sync sync, enum dma_data_direction dir,
+                struct rio_transfer_io *xfer)
+{
+       struct mport_cdev_priv *priv = filp->private_data;
+       unsigned long nr_pages = 0;
+       struct page **page_list = NULL;
+       struct mport_dma_req *req;
+       struct mport_dev *md = priv->md;
+       struct dma_chan *chan;
+       int i, ret;
+       int nents;
+
+       if (xfer->length == 0)
+               return -EINVAL;
+       req = kzalloc(sizeof(*req), GFP_KERNEL);
+       if (!req)
+               return -ENOMEM;
+
+       ret = get_dma_channel(priv);
+       if (ret) {
+               kfree(req);
+               return ret;
+       }
+
+       /*
+        * If parameter loc_addr != NULL, we are transferring data from/to
+        * data buffer allocated in user-space: lock in memory user-space
+        * buffer pages and build an SG table for DMA transfer request
+        *
+        * Otherwise (loc_addr == NULL) contiguous kernel-space buffer is
+        * used for DMA data transfers: build single entry SG table using
+        * offset within the internal buffer specified by handle parameter.
+        */
+       if (xfer->loc_addr) {
+               unsigned long offset;
+               long pinned;
+
+               offset = (unsigned long)xfer->loc_addr & ~PAGE_MASK;
+               nr_pages = PAGE_ALIGN(xfer->length + offset) >> PAGE_SHIFT;
+
+               page_list = kmalloc_array(nr_pages,
+                                         sizeof(*page_list), GFP_KERNEL);
+               if (page_list == NULL) {
+                       ret = -ENOMEM;
+                       goto err_req;
+               }
+
+               down_read(&current->mm->mmap_sem);
+               pinned = get_user_pages(current, current->mm,
+                               (unsigned long)xfer->loc_addr & PAGE_MASK,
+                               nr_pages, dir == DMA_FROM_DEVICE, 0,
+                               page_list, NULL);
+               up_read(&current->mm->mmap_sem);
+
+               if (pinned != nr_pages) {
+                       if (pinned < 0) {
+                               rmcd_error("get_user_pages err=%ld", pinned);
+                               nr_pages = 0;
+                       } else
+                               rmcd_error("pinned %ld out of %ld pages",
+                                          pinned, nr_pages);
+                       ret = -EFAULT;
+                       goto err_pg;
+               }
+
+               ret = sg_alloc_table_from_pages(&req->sgt, page_list, nr_pages,
+                                       offset, xfer->length, GFP_KERNEL);
+               if (ret) {
+                       rmcd_error("sg_alloc_table failed with err=%d", ret);
+                       goto err_pg;
+               }
+
+               req->page_list = page_list;
+               req->nr_pages = nr_pages;
+       } else {
+               dma_addr_t baddr;
+               struct rio_mport_mapping *map;
+
+               baddr = (dma_addr_t)xfer->handle;
+
+               mutex_lock(&md->buf_mutex);
+               list_for_each_entry(map, &md->mappings, node) {
+                       if (baddr >= map->phys_addr &&
+                           baddr < (map->phys_addr + map->size)) {
+                               kref_get(&map->ref);
+                               req->map = map;
+                               break;
+                       }
+               }
+               mutex_unlock(&md->buf_mutex);
+
+               if (req->map == NULL) {
+                       ret = -ENOMEM;
+                       goto err_req;
+               }
+
+               if (xfer->length + xfer->offset > map->size) {
+                       ret = -EINVAL;
+                       goto err_req;
+               }
+
+               ret = sg_alloc_table(&req->sgt, 1, GFP_KERNEL);
+               if (unlikely(ret)) {
+                       rmcd_error("sg_alloc_table failed for internal buf");
+                       goto err_req;
+               }
+
+               sg_set_buf(req->sgt.sgl,
+                          map->virt_addr + (baddr - map->phys_addr) +
+                               xfer->offset, xfer->length);
+       }
+
+       req->dir = dir;
+       req->filp = filp;
+       req->priv = priv;
+       chan = priv->dmach;
+
+       nents = dma_map_sg(chan->device->dev,
+                          req->sgt.sgl, req->sgt.nents, dir);
+       if (nents == -EFAULT) {
+               rmcd_error("Failed to map SG list");
+               return -EFAULT;
+       }
+
+       ret = do_dma_request(req, xfer, sync, nents);
+
+       if (ret >= 0) {
+               if (sync == RIO_TRANSFER_SYNC)
+                       goto sync_out;
+               return ret; /* return ASYNC cookie */
+       }
+
+       if (ret == -ETIMEDOUT || ret == -EINTR) {
+               /*
+                * This can happen only in case of SYNC transfer.
+                * Do not free unfinished request structure immediately.
+                * Place it into pending list and deal with it later
+                */
+               spin_lock(&priv->req_lock);
+               list_add_tail(&req->node, &priv->pend_list);
+               spin_unlock(&priv->req_lock);
+               return ret;
+       }
+
+
+       rmcd_debug(DMA, "do_dma_request failed with err=%d", ret);
+sync_out:
+       dma_unmap_sg(chan->device->dev, req->sgt.sgl, req->sgt.nents, dir);
+       sg_free_table(&req->sgt);
+err_pg:
+       if (page_list) {
+               for (i = 0; i < nr_pages; i++)
+                       put_page(page_list[i]);
+               kfree(page_list);
+       }
+err_req:
+       if (req->map) {
+               mutex_lock(&md->buf_mutex);
+               kref_put(&req->map->ref, mport_release_mapping);
+               mutex_unlock(&md->buf_mutex);
+       }
+       put_dma_channel(priv);
+       kfree(req);
+       return ret;
+}
+
+static int rio_mport_transfer_ioctl(struct file *filp, void __user *arg)
+{
+       struct mport_cdev_priv *priv = filp->private_data;
+       struct rio_transaction transaction;
+       struct rio_transfer_io *transfer;
+       enum dma_data_direction dir;
+       int i, ret = 0;
+
+       if (unlikely(copy_from_user(&transaction, arg, sizeof(transaction))))
+               return -EFAULT;
+
+       if (transaction.count != 1)
+               return -EINVAL;
+
+       if ((transaction.transfer_mode &
+            priv->md->properties.transfer_mode) == 0)
+               return -ENODEV;
+
+       transfer = vmalloc(transaction.count * sizeof(struct rio_transfer_io));
+       if (!transfer)
+               return -ENOMEM;
+
+       if (unlikely(copy_from_user(transfer, transaction.block,
+             transaction.count * sizeof(struct rio_transfer_io)))) {
+               ret = -EFAULT;
+               goto out_free;
+       }
+
+       dir = (transaction.dir == RIO_TRANSFER_DIR_READ) ?
+                                       DMA_FROM_DEVICE : DMA_TO_DEVICE;
+       for (i = 0; i < transaction.count && ret == 0; i++)
+               ret = rio_dma_transfer(filp, transaction.transfer_mode,
+                       transaction.sync, dir, &transfer[i]);
+
+       if (unlikely(copy_to_user(transaction.block, transfer,
+             transaction.count * sizeof(struct rio_transfer_io))))
+               ret = -EFAULT;
+
+out_free:
+       vfree(transfer);
+
+       return ret;
+}
+
+static int rio_mport_wait_for_async_dma(struct file *filp, void __user *arg)
+{
+       struct mport_cdev_priv *priv;
+       struct mport_dev *md;
+       struct rio_async_tx_wait w_param;
+       struct mport_dma_req *req;
+       dma_cookie_t cookie;
+       unsigned long tmo;
+       long wret;
+       int found = 0;
+       int ret;
+
+       priv = (struct mport_cdev_priv *)filp->private_data;
+       md = priv->md;
+
+       if (unlikely(copy_from_user(&w_param, arg, sizeof(w_param))))
+               return -EFAULT;
+
+       cookie = w_param.token;
+       if (w_param.timeout)
+               tmo = msecs_to_jiffies(w_param.timeout);
+       else /* Use default DMA timeout */
+               tmo = msecs_to_jiffies(dma_timeout);
+
+       spin_lock(&priv->req_lock);
+       list_for_each_entry(req, &priv->async_list, node) {
+               if (req->cookie == cookie) {
+                       list_del(&req->node);
+                       found = 1;
+                       break;
+               }
+       }
+       spin_unlock(&priv->req_lock);
+
+       if (!found)
+               return -EAGAIN;
+
+       wret = wait_for_completion_interruptible_timeout(&req->req_comp, tmo);
+
+       if (wret == 0) {
+               /* Timeout on wait occurred */
+               rmcd_error("%s(%d) timed out waiting for ASYNC DMA_%s",
+                      current->comm, task_pid_nr(current),
+                      (req->dir == DMA_FROM_DEVICE)?"READ":"WRITE");
+               ret = -ETIMEDOUT;
+               goto err_tmo;
+       } else if (wret == -ERESTARTSYS) {
+               /* Wait_for_completion was interrupted by a signal but DMA may
+                * be still in progress
+                */
+               rmcd_error("%s(%d) wait for ASYNC DMA_%s was interrupted",
+                       current->comm, task_pid_nr(current),
+                       (req->dir == DMA_FROM_DEVICE)?"READ":"WRITE");
+               ret = -EINTR;
+               goto err_tmo;
+       }
+
+       if (req->status != DMA_COMPLETE) {
+               /* DMA transaction completion signaled with transfer error */
+               rmcd_error("%s(%d) ASYNC DMA_%s completion with status %d",
+                       current->comm, task_pid_nr(current),
+                       (req->dir == DMA_FROM_DEVICE)?"READ":"WRITE",
+                       req->status);
+               ret = -EIO;
+       } else
+               ret = 0;
+
+       if (req->status != DMA_IN_PROGRESS && req->status != DMA_PAUSED)
+               dma_req_free(req);
+
+       return ret;
+
+err_tmo:
+       /* Return request back into async queue */
+       spin_lock(&priv->req_lock);
+       list_add_tail(&req->node, &priv->async_list);
+       spin_unlock(&priv->req_lock);
+       return ret;
+}
+
+static int rio_mport_create_dma_mapping(struct mport_dev *md, struct file *filp,
+                       uint64_t size, struct rio_mport_mapping **mapping)
+{
+       struct rio_mport_mapping *map;
+
+       map = kzalloc(sizeof(struct rio_mport_mapping), GFP_KERNEL);
+       if (map == NULL)
+               return -ENOMEM;
+
+       map->virt_addr = dma_alloc_coherent(md->mport->dev.parent, size,
+                                           &map->phys_addr, GFP_KERNEL);
+       if (map->virt_addr == NULL) {
+               kfree(map);
+               return -ENOMEM;
+       }
+
+       map->dir = MAP_DMA;
+       map->size = size;
+       map->filp = filp;
+       map->md = md;
+       kref_init(&map->ref);
+       mutex_lock(&md->buf_mutex);
+       list_add_tail(&map->node, &md->mappings);
+       mutex_unlock(&md->buf_mutex);
+       *mapping = map;
+
+       return 0;
+}
+
+static int rio_mport_alloc_dma(struct file *filp, void __user *arg)
+{
+       struct mport_cdev_priv *priv = filp->private_data;
+       struct mport_dev *md = priv->md;
+       struct rio_dma_mem map;
+       struct rio_mport_mapping *mapping = NULL;
+       int ret;
+
+       if (unlikely(copy_from_user(&map, arg, sizeof(struct rio_dma_mem))))
+               return -EFAULT;
+
+       ret = rio_mport_create_dma_mapping(md, filp, map.length, &mapping);
+       if (ret)
+               return ret;
+
+       map.dma_handle = mapping->phys_addr;
+
+       if (unlikely(copy_to_user(arg, &map, sizeof(struct rio_dma_mem)))) {
+               mutex_lock(&md->buf_mutex);
+               kref_put(&mapping->ref, mport_release_mapping);
+               mutex_unlock(&md->buf_mutex);
+               return -EFAULT;
+       }
+
+       return 0;
+}
+
+static int rio_mport_free_dma(struct file *filp, void __user *arg)
+{
+       struct mport_cdev_priv *priv = filp->private_data;
+       struct mport_dev *md = priv->md;
+       u64 handle;
+       int ret = -EFAULT;
+       struct rio_mport_mapping *map, *_map;
+
+       if (copy_from_user(&handle, arg, sizeof(u64)))
+               return -EFAULT;
+       rmcd_debug(EXIT, "filp=%p", filp);
+
+       mutex_lock(&md->buf_mutex);
+       list_for_each_entry_safe(map, _map, &md->mappings, node) {
+               if (map->dir == MAP_DMA && map->phys_addr == handle &&
+                   map->filp == filp) {
+                       kref_put(&map->ref, mport_release_mapping);
+                       ret = 0;
+                       break;
+               }
+       }
+       mutex_unlock(&md->buf_mutex);
+
+       if (ret == -EFAULT) {
+               rmcd_debug(DMA, "ERR no matching mapping");
+               return ret;
+       }
+
+       return 0;
+}
+#else
+static int rio_mport_transfer_ioctl(struct file *filp, void *arg)
+{
+       return -ENODEV;
+}
+
+static int rio_mport_wait_for_async_dma(struct file *filp, void __user *arg)
+{
+       return -ENODEV;
+}
+
+static int rio_mport_alloc_dma(struct file *filp, void __user *arg)
+{
+       return -ENODEV;
+}
+
+static int rio_mport_free_dma(struct file *filp, void __user *arg)
+{
+       return -ENODEV;
+}
+#endif /* CONFIG_RAPIDIO_DMA_ENGINE */
+
+/*
+ * Inbound/outbound memory mapping functions
+ */
+
+static int
+rio_mport_create_inbound_mapping(struct mport_dev *md, struct file *filp,
+                               u64 raddr, u32 size,
+                               struct rio_mport_mapping **mapping)
+{
+       struct rio_mport *mport = md->mport;
+       struct rio_mport_mapping *map;
+       int ret;
+
+       map = kzalloc(sizeof(struct rio_mport_mapping), GFP_KERNEL);
+       if (map == NULL)
+               return -ENOMEM;
+
+       map->virt_addr = dma_alloc_coherent(mport->dev.parent, size,
+                                           &map->phys_addr, GFP_KERNEL);
+       if (map->virt_addr == NULL) {
+               ret = -ENOMEM;
+               goto err_dma_alloc;
+       }
+
+       if (raddr == RIO_MAP_ANY_ADDR)
+               raddr = map->phys_addr;
+       ret = rio_map_inb_region(mport, map->phys_addr, raddr, size, 0);
+       if (ret < 0)
+               goto err_map_inb;
+
+       map->dir = MAP_INBOUND;
+       map->rio_addr = raddr;
+       map->size = size;
+       map->filp = filp;
+       map->md = md;
+       kref_init(&map->ref);
+       mutex_lock(&md->buf_mutex);
+       list_add_tail(&map->node, &md->mappings);
+       mutex_unlock(&md->buf_mutex);
+       *mapping = map;
+       return 0;
+
+err_map_inb:
+       dma_free_coherent(mport->dev.parent, size,
+                         map->virt_addr, map->phys_addr);
+err_dma_alloc:
+       kfree(map);
+       return ret;
+}
+
+static int
+rio_mport_get_inbound_mapping(struct mport_dev *md, struct file *filp,
+                             u64 raddr, u32 size,
+                             struct rio_mport_mapping **mapping)
+{
+       struct rio_mport_mapping *map;
+       int err = -ENOMEM;
+
+       if (raddr == RIO_MAP_ANY_ADDR)
+               goto get_new;
+
+       mutex_lock(&md->buf_mutex);
+       list_for_each_entry(map, &md->mappings, node) {
+               if (map->dir != MAP_INBOUND)
+                       continue;
+               if (raddr == map->rio_addr && size == map->size) {
+                       /* allow exact match only */
+                       *mapping = map;
+                       err = 0;
+                       break;
+               } else if (raddr < (map->rio_addr + map->size - 1) &&
+                          (raddr + size) > map->rio_addr) {
+                       err = -EBUSY;
+                       break;
+               }
+       }
+       mutex_unlock(&md->buf_mutex);
+
+       if (err != -ENOMEM)
+               return err;
+get_new:
+       /* not found, create new */
+       return rio_mport_create_inbound_mapping(md, filp, raddr, size, mapping);
+}
+
+static int rio_mport_map_inbound(struct file *filp, void __user *arg)
+{
+       struct mport_cdev_priv *priv = filp->private_data;
+       struct mport_dev *md = priv->md;
+       struct rio_mmap map;
+       struct rio_mport_mapping *mapping = NULL;
+       int ret;
+
+       if (!md->mport->ops->map_inb)
+               return -EPROTONOSUPPORT;
+       if (unlikely(copy_from_user(&map, arg, sizeof(struct rio_mmap))))
+               return -EFAULT;
+
+       rmcd_debug(IBW, "%s filp=%p", dev_name(&priv->md->dev), filp);
+
+       ret = rio_mport_get_inbound_mapping(md, filp, map.rio_addr,
+                                           map.length, &mapping);
+       if (ret)
+               return ret;
+
+       map.handle = mapping->phys_addr;
+       map.rio_addr = mapping->rio_addr;
+
+       if (unlikely(copy_to_user(arg, &map, sizeof(struct rio_mmap)))) {
+               /* Delete mapping if it was created by this request */
+               if (ret == 0 && mapping->filp == filp) {
+                       mutex_lock(&md->buf_mutex);
+                       kref_put(&mapping->ref, mport_release_mapping);
+                       mutex_unlock(&md->buf_mutex);
+               }
+               return -EFAULT;
+       }
+
+       return 0;
+}
+
+/*
+ * rio_mport_inbound_free() - unmap from RapidIO address space and free
+ *                    previously allocated inbound DMA coherent buffer
+ * @priv: driver private data
+ * @arg:  buffer handle returned by allocation routine
+ */
+static int rio_mport_inbound_free(struct file *filp, void __user *arg)
+{
+       struct mport_cdev_priv *priv = filp->private_data;
+       struct mport_dev *md = priv->md;
+       u64 handle;
+       struct rio_mport_mapping *map, *_map;
+
+       rmcd_debug(IBW, "%s filp=%p", dev_name(&priv->md->dev), filp);
+
+       if (!md->mport->ops->unmap_inb)
+               return -EPROTONOSUPPORT;
+
+       if (copy_from_user(&handle, arg, sizeof(u64)))
+               return -EFAULT;
+
+       mutex_lock(&md->buf_mutex);
+       list_for_each_entry_safe(map, _map, &md->mappings, node) {
+               if (map->dir == MAP_INBOUND && map->phys_addr == handle) {
+                       if (map->filp == filp) {
+                               map->filp = NULL;
+                               kref_put(&map->ref, mport_release_mapping);
+                       }
+                       break;
+               }
+       }
+       mutex_unlock(&md->buf_mutex);
+
+       return 0;
+}
+
+/*
+ * maint_port_idx_get() - Get the port index of the mport instance
+ * @priv: driver private data
+ * @arg:  port index
+ */
+static int maint_port_idx_get(struct mport_cdev_priv *priv, void __user *arg)
+{
+       struct mport_dev *md = priv->md;
+       uint32_t port_idx = md->mport->index;
+
+       rmcd_debug(MPORT, "port_index=%d", port_idx);
+
+       if (copy_to_user(arg, &port_idx, sizeof(port_idx)))
+               return -EFAULT;
+
+       return 0;
+}
+
+static int rio_mport_add_event(struct mport_cdev_priv *priv,
+                              struct rio_event *event)
+{
+       int overflow;
+
+       if (!(priv->event_mask & event->header))
+               return -EACCES;
+
+       spin_lock(&priv->fifo_lock);
+       overflow = kfifo_avail(&priv->event_fifo) < sizeof(*event)
+               || kfifo_in(&priv->event_fifo, (unsigned char *)event,
+                       sizeof(*event)) != sizeof(*event);
+       spin_unlock(&priv->fifo_lock);
+
+       wake_up_interruptible(&priv->event_rx_wait);
+
+       if (overflow) {
+               dev_warn(&priv->md->dev, DRV_NAME ": event fifo overflow\n");
+               return -EBUSY;
+       }
+
+       return 0;
+}
+
+static void rio_mport_doorbell_handler(struct rio_mport *mport, void *dev_id,
+                                      u16 src, u16 dst, u16 info)
+{
+       struct mport_dev *data = dev_id;
+       struct mport_cdev_priv *priv;
+       struct rio_mport_db_filter *db_filter;
+       struct rio_event event;
+       int handled;
+
+       event.header = RIO_DOORBELL;
+       event.u.doorbell.rioid = src;
+       event.u.doorbell.payload = info;
+
+       handled = 0;
+       spin_lock(&data->db_lock);
+       list_for_each_entry(db_filter, &data->doorbells, data_node) {
+               if (((db_filter->filter.rioid == 0xffffffff ||
+                     db_filter->filter.rioid == src)) &&
+                     info >= db_filter->filter.low &&
+                     info <= db_filter->filter.high) {
+                       priv = db_filter->priv;
+                       rio_mport_add_event(priv, &event);
+                       handled = 1;
+               }
+       }
+       spin_unlock(&data->db_lock);
+
+       if (!handled)
+               dev_warn(&data->dev,
+                       "%s: spurious DB received from 0x%x, info=0x%04x\n",
+                       __func__, src, info);
+}
+
+static int rio_mport_add_db_filter(struct mport_cdev_priv *priv,
+                                  void __user *arg)
+{
+       struct mport_dev *md = priv->md;
+       struct rio_mport_db_filter *db_filter;
+       struct rio_doorbell_filter filter;
+       unsigned long flags;
+       int ret;
+
+       if (copy_from_user(&filter, arg, sizeof(filter)))
+               return -EFAULT;
+
+       if (filter.low > filter.high)
+               return -EINVAL;
+
+       ret = rio_request_inb_dbell(md->mport, md, filter.low, filter.high,
+                                   rio_mport_doorbell_handler);
+       if (ret) {
+               rmcd_error("%s failed to register IBDB, err=%d",
+                          dev_name(&md->dev), ret);
+               return ret;
+       }
+
+       db_filter = kzalloc(sizeof(*db_filter), GFP_KERNEL);
+       if (db_filter == NULL) {
+               rio_release_inb_dbell(md->mport, filter.low, filter.high);
+               return -ENOMEM;
+       }
+
+       db_filter->filter = filter;
+       db_filter->priv = priv;
+       spin_lock_irqsave(&md->db_lock, flags);
+       list_add_tail(&db_filter->priv_node, &priv->db_filters);
+       list_add_tail(&db_filter->data_node, &md->doorbells);
+       spin_unlock_irqrestore(&md->db_lock, flags);
+
+       return 0;
+}
+
+static void rio_mport_delete_db_filter(struct rio_mport_db_filter *db_filter)
+{
+       list_del(&db_filter->data_node);
+       list_del(&db_filter->priv_node);
+       kfree(db_filter);
+}
+
+static int rio_mport_remove_db_filter(struct mport_cdev_priv *priv,
+                                     void __user *arg)
+{
+       struct rio_mport_db_filter *db_filter;
+       struct rio_doorbell_filter filter;
+       unsigned long flags;
+       int ret = -EINVAL;
+
+       if (copy_from_user(&filter, arg, sizeof(filter)))
+               return -EFAULT;
+
+       spin_lock_irqsave(&priv->md->db_lock, flags);
+       list_for_each_entry(db_filter, &priv->db_filters, priv_node) {
+               if (db_filter->filter.rioid == filter.rioid &&
+                   db_filter->filter.low == filter.low &&
+                   db_filter->filter.high == filter.high) {
+                       rio_mport_delete_db_filter(db_filter);
+                       ret = 0;
+                       break;
+               }
+       }
+       spin_unlock_irqrestore(&priv->md->db_lock, flags);
+
+       if (!ret)
+               rio_release_inb_dbell(priv->md->mport, filter.low, filter.high);
+
+       return ret;
+}
+
+static int rio_mport_match_pw(union rio_pw_msg *msg,
+                             struct rio_pw_filter *filter)
+{
+       if ((msg->em.comptag & filter->mask) < filter->low ||
+               (msg->em.comptag & filter->mask) > filter->high)
+               return 0;
+       return 1;
+}
+
+static int rio_mport_pw_handler(struct rio_mport *mport, void *context,
+                               union rio_pw_msg *msg, int step)
+{
+       struct mport_dev *md = context;
+       struct mport_cdev_priv *priv;
+       struct rio_mport_pw_filter *pw_filter;
+       struct rio_event event;
+       int handled;
+
+       event.header = RIO_PORTWRITE;
+       memcpy(event.u.portwrite.payload, msg->raw, RIO_PW_MSG_SIZE);
+
+       handled = 0;
+       spin_lock(&md->pw_lock);
+       list_for_each_entry(pw_filter, &md->portwrites, md_node) {
+               if (rio_mport_match_pw(msg, &pw_filter->filter)) {
+                       priv = pw_filter->priv;
+                       rio_mport_add_event(priv, &event);
+                       handled = 1;
+               }
+       }
+       spin_unlock(&md->pw_lock);
+
+       if (!handled) {
+               printk_ratelimited(KERN_WARNING DRV_NAME
+                       ": mport%d received spurious PW from 0x%08x\n",
+                       mport->id, msg->em.comptag);
+       }
+
+       return 0;
+}
+
+static int rio_mport_add_pw_filter(struct mport_cdev_priv *priv,
+                                  void __user *arg)
+{
+       struct mport_dev *md = priv->md;
+       struct rio_mport_pw_filter *pw_filter;
+       struct rio_pw_filter filter;
+       unsigned long flags;
+       int hadd = 0;
+
+       if (copy_from_user(&filter, arg, sizeof(filter)))
+               return -EFAULT;
+
+       pw_filter = kzalloc(sizeof(*pw_filter), GFP_KERNEL);
+       if (pw_filter == NULL)
+               return -ENOMEM;
+
+       pw_filter->filter = filter;
+       pw_filter->priv = priv;
+       spin_lock_irqsave(&md->pw_lock, flags);
+       if (list_empty(&md->portwrites))
+               hadd = 1;
+       list_add_tail(&pw_filter->priv_node, &priv->pw_filters);
+       list_add_tail(&pw_filter->md_node, &md->portwrites);
+       spin_unlock_irqrestore(&md->pw_lock, flags);
+
+       if (hadd) {
+               int ret;
+
+               ret = rio_add_mport_pw_handler(md->mport, md,
+                                              rio_mport_pw_handler);
+               if (ret) {
+                       dev_err(&md->dev,
+                               "%s: failed to add IB_PW handler, err=%d\n",
+                               __func__, ret);
+                       return ret;
+               }
+               rio_pw_enable(md->mport, 1);
+       }
+
+       return 0;
+}
+
+static void rio_mport_delete_pw_filter(struct rio_mport_pw_filter *pw_filter)
+{
+       list_del(&pw_filter->md_node);
+       list_del(&pw_filter->priv_node);
+       kfree(pw_filter);
+}
+
+static int rio_mport_match_pw_filter(struct rio_pw_filter *a,
+                                    struct rio_pw_filter *b)
+{
+       if ((a->mask == b->mask) && (a->low == b->low) && (a->high == b->high))
+               return 1;
+       return 0;
+}
+
+static int rio_mport_remove_pw_filter(struct mport_cdev_priv *priv,
+                                     void __user *arg)
+{
+       struct mport_dev *md = priv->md;
+       struct rio_mport_pw_filter *pw_filter;
+       struct rio_pw_filter filter;
+       unsigned long flags;
+       int ret = -EINVAL;
+       int hdel = 0;
+
+       if (copy_from_user(&filter, arg, sizeof(filter)))
+               return -EFAULT;
+
+       spin_lock_irqsave(&md->pw_lock, flags);
+       list_for_each_entry(pw_filter, &priv->pw_filters, priv_node) {
+               if (rio_mport_match_pw_filter(&pw_filter->filter, &filter)) {
+                       rio_mport_delete_pw_filter(pw_filter);
+                       ret = 0;
+                       break;
+               }
+       }
+
+       if (list_empty(&md->portwrites))
+               hdel = 1;
+       spin_unlock_irqrestore(&md->pw_lock, flags);
+
+       if (hdel) {
+               rio_del_mport_pw_handler(md->mport, priv->md,
+                                        rio_mport_pw_handler);
+               rio_pw_enable(md->mport, 0);
+       }
+
+       return ret;
+}
+
+/*
+ * rio_release_dev - release routine for kernel RIO device object
+ * @dev: kernel device object associated with a RIO device structure
+ *
+ * Frees a RIO device struct associated a RIO device struct.
+ * The RIO device struct is freed.
+ */
+static void rio_release_dev(struct device *dev)
+{
+       struct rio_dev *rdev;
+
+       rdev = to_rio_dev(dev);
+       pr_info(DRV_PREFIX "%s: %s\n", __func__, rio_name(rdev));
+       kfree(rdev);
+}
+
+
+static void rio_release_net(struct device *dev)
+{
+       struct rio_net *net;
+
+       net = to_rio_net(dev);
+       rmcd_debug(RDEV, "net_%d", net->id);
+       kfree(net);
+}
+
+
+/*
+ * rio_mport_add_riodev - creates a kernel RIO device object
+ *
+ * Allocates a RIO device data structure and initializes required fields based
+ * on device's configuration space contents.
+ * If the device has switch capabilities, then a switch specific portion is
+ * allocated and configured.
+ */
+static int rio_mport_add_riodev(struct mport_cdev_priv *priv,
+                                  void __user *arg)
+{
+       struct mport_dev *md = priv->md;
+       struct rio_rdev_info dev_info;
+       struct rio_dev *rdev;
+       struct rio_switch *rswitch = NULL;
+       struct rio_mport *mport;
+       size_t size;
+       u32 rval;
+       u32 swpinfo = 0;
+       u16 destid;
+       u8 hopcount;
+       int err;
+
+       if (copy_from_user(&dev_info, arg, sizeof(dev_info)))
+               return -EFAULT;
+
+       rmcd_debug(RDEV, "name:%s ct:0x%x did:0x%x hc:0x%x", dev_info.name,
+                  dev_info.comptag, dev_info.destid, dev_info.hopcount);
+
+       if (bus_find_device_by_name(&rio_bus_type, NULL, dev_info.name)) {
+               rmcd_debug(RDEV, "device %s already exists", dev_info.name);
+               return -EEXIST;
+       }
+
+       size = sizeof(struct rio_dev);
+       mport = md->mport;
+       destid = (u16)dev_info.destid;
+       hopcount = (u8)dev_info.hopcount;
+
+       if (rio_mport_read_config_32(mport, destid, hopcount,
+                                    RIO_PEF_CAR, &rval))
+               return -EIO;
+
+       if (rval & RIO_PEF_SWITCH) {
+               rio_mport_read_config_32(mport, destid, hopcount,
+                                        RIO_SWP_INFO_CAR, &swpinfo);
+               size += (RIO_GET_TOTAL_PORTS(swpinfo) *
+                        sizeof(rswitch->nextdev[0])) + sizeof(*rswitch);
+       }
+
+       rdev = kzalloc(size, GFP_KERNEL);
+       if (rdev == NULL)
+               return -ENOMEM;
+
+       if (mport->net == NULL) {
+               struct rio_net *net;
+
+               net = rio_alloc_net(mport);
+               if (!net) {
+                       err = -ENOMEM;
+                       rmcd_debug(RDEV, "failed to allocate net object");
+                       goto cleanup;
+               }
+
+               net->id = mport->id;
+               net->hport = mport;
+               dev_set_name(&net->dev, "rnet_%d", net->id);
+               net->dev.parent = &mport->dev;
+               net->dev.release = rio_release_net;
+               err = rio_add_net(net);
+               if (err) {
+                       rmcd_debug(RDEV, "failed to register net, err=%d", err);
+                       kfree(net);
+                       goto cleanup;
+               }
+       }
+
+       rdev->net = mport->net;
+       rdev->pef = rval;
+       rdev->swpinfo = swpinfo;
+       rio_mport_read_config_32(mport, destid, hopcount,
+                                RIO_DEV_ID_CAR, &rval);
+       rdev->did = rval >> 16;
+       rdev->vid = rval & 0xffff;
+       rio_mport_read_config_32(mport, destid, hopcount, RIO_DEV_INFO_CAR,
+                                &rdev->device_rev);
+       rio_mport_read_config_32(mport, destid, hopcount, RIO_ASM_ID_CAR,
+                                &rval);
+       rdev->asm_did = rval >> 16;
+       rdev->asm_vid = rval & 0xffff;
+       rio_mport_read_config_32(mport, destid, hopcount, RIO_ASM_INFO_CAR,
+                                &rval);
+       rdev->asm_rev = rval >> 16;
+
+       if (rdev->pef & RIO_PEF_EXT_FEATURES) {
+               rdev->efptr = rval & 0xffff;
+               rdev->phys_efptr = rio_mport_get_physefb(mport, 0, destid,
+                                                        hopcount);
+
+               rdev->em_efptr = rio_mport_get_feature(mport, 0, destid,
+                                               hopcount, RIO_EFB_ERR_MGMNT);
+       }
+
+       rio_mport_read_config_32(mport, destid, hopcount, RIO_SRC_OPS_CAR,
+                                &rdev->src_ops);
+       rio_mport_read_config_32(mport, destid, hopcount, RIO_DST_OPS_CAR,
+                                &rdev->dst_ops);
+
+       rdev->comp_tag = dev_info.comptag;
+       rdev->destid = destid;
+       /* hopcount is stored as specified by a caller, regardles of EP or SW */
+       rdev->hopcount = hopcount;
+
+       if (rdev->pef & RIO_PEF_SWITCH) {
+               rswitch = rdev->rswitch;
+               rswitch->route_table = NULL;
+       }
+
+       if (strlen(dev_info.name))
+               dev_set_name(&rdev->dev, "%s", dev_info.name);
+       else if (rdev->pef & RIO_PEF_SWITCH)
+               dev_set_name(&rdev->dev, "%02x:s:%04x", mport->id,
+                            rdev->comp_tag & RIO_CTAG_UDEVID);
+       else
+               dev_set_name(&rdev->dev, "%02x:e:%04x", mport->id,
+                            rdev->comp_tag & RIO_CTAG_UDEVID);
+
+       INIT_LIST_HEAD(&rdev->net_list);
+       rdev->dev.parent = &mport->net->dev;
+       rio_attach_device(rdev);
+       rdev->dev.release = rio_release_dev;
+
+       if (rdev->dst_ops & RIO_DST_OPS_DOORBELL)
+               rio_init_dbell_res(&rdev->riores[RIO_DOORBELL_RESOURCE],
+                                  0, 0xffff);
+       err = rio_add_device(rdev);
+       if (err)
+               goto cleanup;
+       rio_dev_get(rdev);
+
+       return 0;
+cleanup:
+       kfree(rdev);
+       return err;
+}
+
+static int rio_mport_del_riodev(struct mport_cdev_priv *priv, void __user *arg)
+{
+       struct rio_rdev_info dev_info;
+       struct rio_dev *rdev = NULL;
+       struct device  *dev;
+       struct rio_mport *mport;
+       struct rio_net *net;
+
+       if (copy_from_user(&dev_info, arg, sizeof(dev_info)))
+               return -EFAULT;
+
+       mport = priv->md->mport;
+
+       /* If device name is specified, removal by name has priority */
+       if (strlen(dev_info.name)) {
+               dev = bus_find_device_by_name(&rio_bus_type, NULL,
+                                             dev_info.name);
+               if (dev)
+                       rdev = to_rio_dev(dev);
+       } else {
+               do {
+                       rdev = rio_get_comptag(dev_info.comptag, rdev);
+                       if (rdev && rdev->dev.parent == &mport->net->dev &&
+                           rdev->destid == (u16)dev_info.destid &&
+                           rdev->hopcount == (u8)dev_info.hopcount)
+                               break;
+               } while (rdev);
+       }
+
+       if (!rdev) {
+               rmcd_debug(RDEV,
+                       "device name:%s ct:0x%x did:0x%x hc:0x%x not found",
+                       dev_info.name, dev_info.comptag, dev_info.destid,
+                       dev_info.hopcount);
+               return -ENODEV;
+       }
+
+       net = rdev->net;
+       rio_dev_put(rdev);
+       rio_del_device(rdev, RIO_DEVICE_SHUTDOWN);
+
+       if (list_empty(&net->devices)) {
+               rio_free_net(net);
+               mport->net = NULL;
+       }
+
+       return 0;
+}
+
+/*
+ * Mport cdev management
+ */
+
+/*
+ * mport_cdev_open() - Open character device (mport)
+ */
+static int mport_cdev_open(struct inode *inode, struct file *filp)
+{
+       int ret;
+       int minor = iminor(inode);
+       struct mport_dev *chdev;
+       struct mport_cdev_priv *priv;
+
+       /* Test for valid device */
+       if (minor >= RIO_MAX_MPORTS) {
+               rmcd_error("Invalid minor device number");
+               return -EINVAL;
+       }
+
+       chdev = container_of(inode->i_cdev, struct mport_dev, cdev);
+
+       rmcd_debug(INIT, "%s filp=%p", dev_name(&chdev->dev), filp);
+
+       if (atomic_read(&chdev->active) == 0)
+               return -ENODEV;
+
+       get_device(&chdev->dev);
+
+       priv = kzalloc(sizeof(*priv), GFP_KERNEL);
+       if (!priv) {
+               put_device(&chdev->dev);
+               return -ENOMEM;
+       }
+
+       priv->md = chdev;
+
+       mutex_lock(&chdev->file_mutex);
+       list_add_tail(&priv->list, &chdev->file_list);
+       mutex_unlock(&chdev->file_mutex);
+
+       INIT_LIST_HEAD(&priv->db_filters);
+       INIT_LIST_HEAD(&priv->pw_filters);
+       spin_lock_init(&priv->fifo_lock);
+       init_waitqueue_head(&priv->event_rx_wait);
+       ret = kfifo_alloc(&priv->event_fifo,
+                         sizeof(struct rio_event) * MPORT_EVENT_DEPTH,
+                         GFP_KERNEL);
+       if (ret < 0) {
+               dev_err(&chdev->dev, DRV_NAME ": kfifo_alloc failed\n");
+               ret = -ENOMEM;
+               goto err_fifo;
+       }
+
+#ifdef CONFIG_RAPIDIO_DMA_ENGINE
+       INIT_LIST_HEAD(&priv->async_list);
+       INIT_LIST_HEAD(&priv->pend_list);
+       spin_lock_init(&priv->req_lock);
+       mutex_init(&priv->dma_lock);
+#endif
+
+       filp->private_data = priv;
+       goto out;
+err_fifo:
+       kfree(priv);
+out:
+       return ret;
+}
+
+static int mport_cdev_fasync(int fd, struct file *filp, int mode)
+{
+       struct mport_cdev_priv *priv = filp->private_data;
+
+       return fasync_helper(fd, filp, mode, &priv->async_queue);
+}
+
+#ifdef CONFIG_RAPIDIO_DMA_ENGINE
+static void mport_cdev_release_dma(struct file *filp)
+{
+       struct mport_cdev_priv *priv = filp->private_data;
+       struct mport_dev *md;
+       struct mport_dma_req *req, *req_next;
+       unsigned long tmo = msecs_to_jiffies(dma_timeout);
+       long wret;
+       LIST_HEAD(list);
+
+       rmcd_debug(EXIT, "from filp=%p %s(%d)",
+                  filp, current->comm, task_pid_nr(current));
+
+       if (!priv->dmach) {
+               rmcd_debug(EXIT, "No DMA channel for filp=%p", filp);
+               return;
+       }
+
+       md = priv->md;
+
+       flush_workqueue(dma_wq);
+
+       spin_lock(&priv->req_lock);
+       if (!list_empty(&priv->async_list)) {
+               rmcd_debug(EXIT, "async list not empty filp=%p %s(%d)",
+                          filp, current->comm, task_pid_nr(current));
+               list_splice_init(&priv->async_list, &list);
+       }
+       spin_unlock(&priv->req_lock);
+
+       if (!list_empty(&list)) {
+               rmcd_debug(EXIT, "temp list not empty");
+               list_for_each_entry_safe(req, req_next, &list, node) {
+                       rmcd_debug(EXIT, "free req->filp=%p cookie=%d compl=%s",
+                                  req->filp, req->cookie,
+                                  completion_done(&req->req_comp)?"yes":"no");
+                       list_del(&req->node);
+                       dma_req_free(req);
+               }
+       }
+
+       if (!list_empty(&priv->pend_list)) {
+               rmcd_debug(EXIT, "Free pending DMA requests for filp=%p %s(%d)",
+                          filp, current->comm, task_pid_nr(current));
+               list_for_each_entry_safe(req,
+                                        req_next, &priv->pend_list, node) {
+                       rmcd_debug(EXIT, "free req->filp=%p cookie=%d compl=%s",
+                                  req->filp, req->cookie,
+                                  completion_done(&req->req_comp)?"yes":"no");
+                       list_del(&req->node);
+                       dma_req_free(req);
+               }
+       }
+
+       put_dma_channel(priv);
+       wret = wait_for_completion_interruptible_timeout(&priv->comp, tmo);
+
+       if (wret <= 0) {
+               rmcd_error("%s(%d) failed waiting for DMA release err=%ld",
+                       current->comm, task_pid_nr(current), wret);
+       }
+
+       spin_lock(&priv->req_lock);
+
+       if (!list_empty(&priv->pend_list)) {
+               rmcd_debug(EXIT, "ATTN: pending DMA requests, filp=%p %s(%d)",
+                          filp, current->comm, task_pid_nr(current));
+       }
+
+       spin_unlock(&priv->req_lock);
+
+       if (priv->dmach != priv->md->dma_chan) {
+               rmcd_debug(EXIT, "Release DMA channel for filp=%p %s(%d)",
+                          filp, current->comm, task_pid_nr(current));
+               rio_release_dma(priv->dmach);
+       } else {
+               rmcd_debug(EXIT, "Adjust default DMA channel refcount");
+               kref_put(&md->dma_ref, mport_release_def_dma);
+       }
+
+       priv->dmach = NULL;
+}
+#else
+#define mport_cdev_release_dma(priv) do {} while (0)
+#endif
+
+/*
+ * mport_cdev_release() - Release character device
+ */
+static int mport_cdev_release(struct inode *inode, struct file *filp)
+{
+       struct mport_cdev_priv *priv = filp->private_data;
+       struct mport_dev *chdev;
+       struct rio_mport_pw_filter *pw_filter, *pw_filter_next;
+       struct rio_mport_db_filter *db_filter, *db_filter_next;
+       struct rio_mport_mapping *map, *_map;
+       unsigned long flags;
+
+       rmcd_debug(EXIT, "%s filp=%p", dev_name(&priv->md->dev), filp);
+
+       chdev = priv->md;
+       mport_cdev_release_dma(filp);
+
+       priv->event_mask = 0;
+
+       spin_lock_irqsave(&chdev->pw_lock, flags);
+       if (!list_empty(&priv->pw_filters)) {
+               list_for_each_entry_safe(pw_filter, pw_filter_next,
+                                        &priv->pw_filters, priv_node)
+                       rio_mport_delete_pw_filter(pw_filter);
+       }
+       spin_unlock_irqrestore(&chdev->pw_lock, flags);
+
+       spin_lock_irqsave(&chdev->db_lock, flags);
+       list_for_each_entry_safe(db_filter, db_filter_next,
+                                &priv->db_filters, priv_node) {
+               rio_mport_delete_db_filter(db_filter);
+       }
+       spin_unlock_irqrestore(&chdev->db_lock, flags);
+
+       kfifo_free(&priv->event_fifo);
+
+       mutex_lock(&chdev->buf_mutex);
+       list_for_each_entry_safe(map, _map, &chdev->mappings, node) {
+               if (map->filp == filp) {
+                       rmcd_debug(EXIT, "release mapping %p filp=%p",
+                                  map->virt_addr, filp);
+                       kref_put(&map->ref, mport_release_mapping);
+               }
+       }
+       mutex_unlock(&chdev->buf_mutex);
+
+       mport_cdev_fasync(-1, filp, 0);
+       filp->private_data = NULL;
+       mutex_lock(&chdev->file_mutex);
+       list_del(&priv->list);
+       mutex_unlock(&chdev->file_mutex);
+       put_device(&chdev->dev);
+       kfree(priv);
+       return 0;
+}
+
+/*
+ * mport_cdev_ioctl() - IOCTLs for character device
+ */
+static long mport_cdev_ioctl(struct file *filp,
+               unsigned int cmd, unsigned long arg)
+{
+       int err = -EINVAL;
+       struct mport_cdev_priv *data = filp->private_data;
+       struct mport_dev *md = data->md;
+
+       if (atomic_read(&md->active) == 0)
+               return -ENODEV;
+
+       switch (cmd) {
+       case RIO_MPORT_MAINT_READ_LOCAL:
+               return rio_mport_maint_rd(data, (void __user *)arg, 1);
+       case RIO_MPORT_MAINT_WRITE_LOCAL:
+               return rio_mport_maint_wr(data, (void __user *)arg, 1);
+       case RIO_MPORT_MAINT_READ_REMOTE:
+               return rio_mport_maint_rd(data, (void __user *)arg, 0);
+       case RIO_MPORT_MAINT_WRITE_REMOTE:
+               return rio_mport_maint_wr(data, (void __user *)arg, 0);
+       case RIO_MPORT_MAINT_HDID_SET:
+               return maint_hdid_set(data, (void __user *)arg);
+       case RIO_MPORT_MAINT_COMPTAG_SET:
+               return maint_comptag_set(data, (void __user *)arg);
+       case RIO_MPORT_MAINT_PORT_IDX_GET:
+               return maint_port_idx_get(data, (void __user *)arg);
+       case RIO_MPORT_GET_PROPERTIES:
+               md->properties.hdid = md->mport->host_deviceid;
+               if (copy_to_user((void __user *)arg, &(data->md->properties),
+                                sizeof(data->md->properties)))
+                       return -EFAULT;
+               return 0;
+       case RIO_ENABLE_DOORBELL_RANGE:
+               return rio_mport_add_db_filter(data, (void __user *)arg);
+       case RIO_DISABLE_DOORBELL_RANGE:
+               return rio_mport_remove_db_filter(data, (void __user *)arg);
+       case RIO_ENABLE_PORTWRITE_RANGE:
+               return rio_mport_add_pw_filter(data, (void __user *)arg);
+       case RIO_DISABLE_PORTWRITE_RANGE:
+               return rio_mport_remove_pw_filter(data, (void __user *)arg);
+       case RIO_SET_EVENT_MASK:
+               data->event_mask = arg;
+               return 0;
+       case RIO_GET_EVENT_MASK:
+               if (copy_to_user((void __user *)arg, &data->event_mask,
+                                   sizeof(data->event_mask)))
+                       return -EFAULT;
+               return 0;
+       case RIO_MAP_OUTBOUND:
+               return rio_mport_obw_map(filp, (void __user *)arg);
+       case RIO_MAP_INBOUND:
+               return rio_mport_map_inbound(filp, (void __user *)arg);
+       case RIO_UNMAP_OUTBOUND:
+               return rio_mport_obw_free(filp, (void __user *)arg);
+       case RIO_UNMAP_INBOUND:
+               return rio_mport_inbound_free(filp, (void __user *)arg);
+       case RIO_ALLOC_DMA:
+               return rio_mport_alloc_dma(filp, (void __user *)arg);
+       case RIO_FREE_DMA:
+               return rio_mport_free_dma(filp, (void __user *)arg);
+       case RIO_WAIT_FOR_ASYNC:
+               return rio_mport_wait_for_async_dma(filp, (void __user *)arg);
+       case RIO_TRANSFER:
+               return rio_mport_transfer_ioctl(filp, (void __user *)arg);
+       case RIO_DEV_ADD:
+               return rio_mport_add_riodev(data, (void __user *)arg);
+       case RIO_DEV_DEL:
+               return rio_mport_del_riodev(data, (void __user *)arg);
+       default:
+               break;
+       }
+
+       return err;
+}
+
+/*
+ * mport_release_mapping - free mapping resources and info structure
+ * @ref: a pointer to the kref within struct rio_mport_mapping
+ *
+ * NOTE: Shall be called while holding buf_mutex.
+ */
+static void mport_release_mapping(struct kref *ref)
+{
+       struct rio_mport_mapping *map =
+                       container_of(ref, struct rio_mport_mapping, ref);
+       struct rio_mport *mport = map->md->mport;
+
+       rmcd_debug(MMAP, "type %d mapping @ %p (phys = %pad) for %s",
+                  map->dir, map->virt_addr,
+                  &map->phys_addr, mport->name);
+
+       list_del(&map->node);
+
+       switch (map->dir) {
+       case MAP_INBOUND:
+               rio_unmap_inb_region(mport, map->phys_addr);
+       case MAP_DMA:
+               dma_free_coherent(mport->dev.parent, map->size,
+                                 map->virt_addr, map->phys_addr);
+               break;
+       case MAP_OUTBOUND:
+               rio_unmap_outb_region(mport, map->rioid, map->rio_addr);
+               break;
+       }
+       kfree(map);
+}
+
+static void mport_mm_open(struct vm_area_struct *vma)
+{
+       struct rio_mport_mapping *map = vma->vm_private_data;
+
+rmcd_debug(MMAP, "0x%pad", &map->phys_addr);
+       kref_get(&map->ref);
+}
+
+static void mport_mm_close(struct vm_area_struct *vma)
+{
+       struct rio_mport_mapping *map = vma->vm_private_data;
+
+rmcd_debug(MMAP, "0x%pad", &map->phys_addr);
+       mutex_lock(&map->md->buf_mutex);
+       kref_put(&map->ref, mport_release_mapping);
+       mutex_unlock(&map->md->buf_mutex);
+}
+
+static const struct vm_operations_struct vm_ops = {
+       .open = mport_mm_open,
+       .close = mport_mm_close,
+};
+
+static int mport_cdev_mmap(struct file *filp, struct vm_area_struct *vma)
+{
+       struct mport_cdev_priv *priv = filp->private_data;
+       struct mport_dev *md;
+       size_t size = vma->vm_end - vma->vm_start;
+       dma_addr_t baddr;
+       unsigned long offset;
+       int found = 0, ret;
+       struct rio_mport_mapping *map;
+
+       rmcd_debug(MMAP, "0x%x bytes at offset 0x%lx",
+                  (unsigned int)size, vma->vm_pgoff);
+
+       md = priv->md;
+       baddr = ((dma_addr_t)vma->vm_pgoff << PAGE_SHIFT);
+
+       mutex_lock(&md->buf_mutex);
+       list_for_each_entry(map, &md->mappings, node) {
+               if (baddr >= map->phys_addr &&
+                   baddr < (map->phys_addr + map->size)) {
+                       found = 1;
+                       break;
+               }
+       }
+       mutex_unlock(&md->buf_mutex);
+
+       if (!found)
+               return -ENOMEM;
+
+       offset = baddr - map->phys_addr;
+
+       if (size + offset > map->size)
+               return -EINVAL;
+
+       vma->vm_pgoff = offset >> PAGE_SHIFT;
+       rmcd_debug(MMAP, "MMAP adjusted offset = 0x%lx", vma->vm_pgoff);
+
+       if (map->dir == MAP_INBOUND || map->dir == MAP_DMA)
+               ret = dma_mmap_coherent(md->mport->dev.parent, vma,
+                               map->virt_addr, map->phys_addr, map->size);
+       else if (map->dir == MAP_OUTBOUND) {
+               vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
+               ret = vm_iomap_memory(vma, map->phys_addr, map->size);
+       } else {
+               rmcd_error("Attempt to mmap unsupported mapping type");
+               ret = -EIO;
+       }
+
+       if (!ret) {
+               vma->vm_private_data = map;
+               vma->vm_ops = &vm_ops;
+               mport_mm_open(vma);
+       } else {
+               rmcd_error("MMAP exit with err=%d", ret);
+       }
+
+       return ret;
+}
+
+static unsigned int mport_cdev_poll(struct file *filp, poll_table *wait)
+{
+       struct mport_cdev_priv *priv = filp->private_data;
+
+       poll_wait(filp, &priv->event_rx_wait, wait);
+       if (kfifo_len(&priv->event_fifo))
+               return POLLIN | POLLRDNORM;
+
+       return 0;
+}
+
+static ssize_t mport_read(struct file *filp, char __user *buf, size_t count,
+                       loff_t *ppos)
+{
+       struct mport_cdev_priv *priv = filp->private_data;
+       int copied;
+       ssize_t ret;
+
+       if (!count)
+               return 0;
+
+       if (kfifo_is_empty(&priv->event_fifo) &&
+           (filp->f_flags & O_NONBLOCK))
+               return -EAGAIN;
+
+       if (count % sizeof(struct rio_event))
+               return -EINVAL;
+
+       ret = wait_event_interruptible(priv->event_rx_wait,
+                                       kfifo_len(&priv->event_fifo) != 0);
+       if (ret)
+               return ret;
+
+       while (ret < count) {
+               if (kfifo_to_user(&priv->event_fifo, buf,
+                     sizeof(struct rio_event), &copied))
+                       return -EFAULT;
+               ret += copied;
+               buf += copied;
+       }
+
+       return ret;
+}
+
+static ssize_t mport_write(struct file *filp, const char __user *buf,
+                        size_t count, loff_t *ppos)
+{
+       struct mport_cdev_priv *priv = filp->private_data;
+       struct rio_mport *mport = priv->md->mport;
+       struct rio_event event;
+       int len, ret;
+
+       if (!count)
+               return 0;
+
+       if (count % sizeof(event))
+               return -EINVAL;
+
+       len = 0;
+       while ((count - len) >= (int)sizeof(event)) {
+               if (copy_from_user(&event, buf, sizeof(event)))
+                       return -EFAULT;
+
+               if (event.header != RIO_DOORBELL)
+                       return -EINVAL;
+
+               ret = rio_mport_send_doorbell(mport,
+                                             (u16)event.u.doorbell.rioid,
+                                             event.u.doorbell.payload);
+               if (ret < 0)
+                       return ret;
+
+               len += sizeof(event);
+               buf += sizeof(event);
+       }
+
+       return len;
+}
+
+static const struct file_operations mport_fops = {
+       .owner          = THIS_MODULE,
+       .open           = mport_cdev_open,
+       .release        = mport_cdev_release,
+       .poll           = mport_cdev_poll,
+       .read           = mport_read,
+       .write          = mport_write,
+       .mmap           = mport_cdev_mmap,
+       .fasync         = mport_cdev_fasync,
+       .unlocked_ioctl = mport_cdev_ioctl
+};
+
+/*
+ * Character device management
+ */
+
+static void mport_device_release(struct device *dev)
+{
+       struct mport_dev *md;
+
+       rmcd_debug(EXIT, "%s", dev_name(dev));
+       md = container_of(dev, struct mport_dev, dev);
+       kfree(md);
+}
+
+/*
+ * mport_cdev_add() - Create mport_dev from rio_mport
+ * @mport:     RapidIO master port
+ */
+static struct mport_dev *mport_cdev_add(struct rio_mport *mport)
+{
+       int ret = 0;
+       struct mport_dev *md;
+       struct rio_mport_attr attr;
+
+       md = kzalloc(sizeof(struct mport_dev), GFP_KERNEL);
+       if (!md) {
+               rmcd_error("Unable allocate a device object");
+               return NULL;
+       }
+
+       md->mport = mport;
+       mutex_init(&md->buf_mutex);
+       mutex_init(&md->file_mutex);
+       INIT_LIST_HEAD(&md->file_list);
+       cdev_init(&md->cdev, &mport_fops);
+       md->cdev.owner = THIS_MODULE;
+       ret = cdev_add(&md->cdev, MKDEV(MAJOR(dev_number), mport->id), 1);
+       if (ret < 0) {
+               kfree(md);
+               rmcd_error("Unable to register a device, err=%d", ret);
+               return NULL;
+       }
+
+       md->dev.devt = md->cdev.dev;
+       md->dev.class = dev_class;
+       md->dev.parent = &mport->dev;
+       md->dev.release = mport_device_release;
+       dev_set_name(&md->dev, DEV_NAME "%d", mport->id);
+       atomic_set(&md->active, 1);
+
+       ret = device_register(&md->dev);
+       if (ret) {
+               rmcd_error("Failed to register mport %d (err=%d)",
+                      mport->id, ret);
+               goto err_cdev;
+       }
+
+       get_device(&md->dev);
+
+       INIT_LIST_HEAD(&md->doorbells);
+       spin_lock_init(&md->db_lock);
+       INIT_LIST_HEAD(&md->portwrites);
+       spin_lock_init(&md->pw_lock);
+       INIT_LIST_HEAD(&md->mappings);
+
+       md->properties.id = mport->id;
+       md->properties.sys_size = mport->sys_size;
+       md->properties.hdid = mport->host_deviceid;
+       md->properties.index = mport->index;
+
+       /* The transfer_mode property will be returned through mport query
+        * interface
+        */
+#ifdef CONFIG_PPC /* for now: only on Freescale's SoCs */
+       md->properties.transfer_mode |= RIO_TRANSFER_MODE_MAPPED;
+#else
+       md->properties.transfer_mode |= RIO_TRANSFER_MODE_TRANSFER;
+#endif
+       ret = rio_query_mport(mport, &attr);
+       if (!ret) {
+               md->properties.flags = attr.flags;
+               md->properties.link_speed = attr.link_speed;
+               md->properties.link_width = attr.link_width;
+               md->properties.dma_max_sge = attr.dma_max_sge;
+               md->properties.dma_max_size = attr.dma_max_size;
+               md->properties.dma_align = attr.dma_align;
+               md->properties.cap_sys_size = 0;
+               md->properties.cap_transfer_mode = 0;
+               md->properties.cap_addr_size = 0;
+       } else
+               pr_info(DRV_PREFIX "Failed to obtain info for %s cdev(%d:%d)\n",
+                       mport->name, MAJOR(dev_number), mport->id);
+
+       mutex_lock(&mport_devs_lock);
+       list_add_tail(&md->node, &mport_devs);
+       mutex_unlock(&mport_devs_lock);
+
+       pr_info(DRV_PREFIX "Added %s cdev(%d:%d)\n",
+               mport->name, MAJOR(dev_number), mport->id);
+
+       return md;
+
+err_cdev:
+       cdev_del(&md->cdev);
+       kfree(md);
+       return NULL;
+}
+
+/*
+ * mport_cdev_terminate_dma() - Stop all active DMA data transfers and release
+ *                              associated DMA channels.
+ */
+static void mport_cdev_terminate_dma(struct mport_dev *md)
+{
+#ifdef CONFIG_RAPIDIO_DMA_ENGINE
+       struct mport_cdev_priv *client;
+
+       rmcd_debug(DMA, "%s", dev_name(&md->dev));
+
+       mutex_lock(&md->file_mutex);
+       list_for_each_entry(client, &md->file_list, list) {
+               if (client->dmach) {
+                       dmaengine_terminate_all(client->dmach);
+                       rio_release_dma(client->dmach);
+               }
+       }
+       mutex_unlock(&md->file_mutex);
+
+       if (md->dma_chan) {
+               dmaengine_terminate_all(md->dma_chan);
+               rio_release_dma(md->dma_chan);
+               md->dma_chan = NULL;
+       }
+#endif
+}
+
+
+/*
+ * mport_cdev_kill_fasync() - Send SIGIO signal to all processes with open
+ *                            mport_cdev files.
+ */
+static int mport_cdev_kill_fasync(struct mport_dev *md)
+{
+       unsigned int files = 0;
+       struct mport_cdev_priv *client;
+
+       mutex_lock(&md->file_mutex);
+       list_for_each_entry(client, &md->file_list, list) {
+               if (client->async_queue)
+                       kill_fasync(&client->async_queue, SIGIO, POLL_HUP);
+               files++;
+       }
+       mutex_unlock(&md->file_mutex);
+       return files;
+}
+
+/*
+ * mport_cdev_remove() - Remove mport character device
+ * @dev:       Mport device to remove
+ */
+static void mport_cdev_remove(struct mport_dev *md)
+{
+       struct rio_mport_mapping *map, *_map;
+
+       rmcd_debug(EXIT, "Remove %s cdev", md->mport->name);
+       atomic_set(&md->active, 0);
+       mport_cdev_terminate_dma(md);
+       rio_del_mport_pw_handler(md->mport, md, rio_mport_pw_handler);
+       cdev_del(&(md->cdev));
+       mport_cdev_kill_fasync(md);
+
+       flush_workqueue(dma_wq);
+
+       /* TODO: do we need to give clients some time to close file
+        * descriptors? Simple wait for XX, or kref?
+        */
+
+       /*
+        * Release DMA buffers allocated for the mport device.
+        * Disable associated inbound Rapidio requests mapping if applicable.
+        */
+       mutex_lock(&md->buf_mutex);
+       list_for_each_entry_safe(map, _map, &md->mappings, node) {
+               kref_put(&map->ref, mport_release_mapping);
+       }
+       mutex_unlock(&md->buf_mutex);
+
+       if (!list_empty(&md->mappings))
+               rmcd_warn("WARNING: %s pending mappings on removal",
+                         md->mport->name);
+
+       rio_release_inb_dbell(md->mport, 0, 0x0fff);
+
+       device_unregister(&md->dev);
+       put_device(&md->dev);
+}
+
+/*
+ * RIO rio_mport_interface driver
+ */
+
+/*
+ * mport_add_mport() - Add rio_mport from LDM device struct
+ * @dev:               Linux device model struct
+ * @class_intf:        Linux class_interface
+ */
+static int mport_add_mport(struct device *dev,
+               struct class_interface *class_intf)
+{
+       struct rio_mport *mport = NULL;
+       struct mport_dev *chdev = NULL;
+
+       mport = to_rio_mport(dev);
+       if (!mport)
+               return -ENODEV;
+
+       chdev = mport_cdev_add(mport);
+       if (!chdev)
+               return -ENODEV;
+
+       return 0;
+}
+
+/*
+ * mport_remove_mport() - Remove rio_mport from global list
+ * TODO remove device from global mport_dev list
+ */
+static void mport_remove_mport(struct device *dev,
+               struct class_interface *class_intf)
+{
+       struct rio_mport *mport = NULL;
+       struct mport_dev *chdev;
+       int found = 0;
+
+       mport = to_rio_mport(dev);
+       rmcd_debug(EXIT, "Remove %s", mport->name);
+
+       mutex_lock(&mport_devs_lock);
+       list_for_each_entry(chdev, &mport_devs, node) {
+               if (chdev->mport->id == mport->id) {
+                       atomic_set(&chdev->active, 0);
+                       list_del(&chdev->node);
+                       found = 1;
+                       break;
+               }
+       }
+       mutex_unlock(&mport_devs_lock);
+
+       if (found)
+               mport_cdev_remove(chdev);
+}
+
+/* the rio_mport_interface is used to handle local mport devices */
+static struct class_interface rio_mport_interface __refdata = {
+       .class          = &rio_mport_class,
+       .add_dev        = mport_add_mport,
+       .remove_dev     = mport_remove_mport,
+};
+
+/*
+ * Linux kernel module
+ */
+
+/*
+ * mport_init - Driver module loading
+ */
+static int __init mport_init(void)
+{
+       int ret;
+
+       /* Create device class needed by udev */
+       dev_class = class_create(THIS_MODULE, DRV_NAME);
+       if (!dev_class) {
+               rmcd_error("Unable to create " DRV_NAME " class");
+               return -EINVAL;
+       }
+
+       ret = alloc_chrdev_region(&dev_number, 0, RIO_MAX_MPORTS, DRV_NAME);
+       if (ret < 0)
+               goto err_chr;
+
+       rmcd_debug(INIT, "Registered class with major=%d", MAJOR(dev_number));
+
+       /* Register to rio_mport_interface */
+       ret = class_interface_register(&rio_mport_interface);
+       if (ret) {
+               rmcd_error("class_interface_register() failed, err=%d", ret);
+               goto err_cli;
+       }
+
+       dma_wq = create_singlethread_workqueue("dma_wq");
+       if (!dma_wq) {
+               rmcd_error("failed to create DMA work queue");
+               ret = -ENOMEM;
+               goto err_wq;
+       }
+
+       return 0;
+
+err_wq:
+       class_interface_unregister(&rio_mport_interface);
+err_cli:
+       unregister_chrdev_region(dev_number, RIO_MAX_MPORTS);
+err_chr:
+       class_destroy(dev_class);
+       return ret;
+}
+
+/**
+ * mport_exit - Driver module unloading
+ */
+static void __exit mport_exit(void)
+{
+       class_interface_unregister(&rio_mport_interface);
+       class_destroy(dev_class);
+       unregister_chrdev_region(dev_number, RIO_MAX_MPORTS);
+       destroy_workqueue(dma_wq);
+}
+
+module_init(mport_init);
+module_exit(mport_exit);
diff --git a/include/linux/rio_mport_cdev.h b/include/linux/rio_mport_cdev.h
new file mode 100644 (file)
index 0000000..b65d19d
--- /dev/null
@@ -0,0 +1,271 @@
+/*
+ * Copyright (c) 2015-2016, Integrated Device Technology Inc.
+ * Copyright (c) 2015, Prodrive Technologies
+ * Copyright (c) 2015, Texas Instruments Incorporated
+ * Copyright (c) 2015, RapidIO Trade Association
+ * All rights reserved.
+ *
+ * This software is available to you under a choice of one of two licenses.
+ * You may choose to be licensed under the terms of the GNU General Public
+ * License(GPL) Version 2, or the BSD-3 Clause license below:
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ * this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ * this list of conditions and the following disclaimer in the documentation
+ * and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its contributors
+ * may be used to endorse or promote products derived from this software without
+ * specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
+ * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR
+ * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+ * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
+ * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+ * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
+ * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+ * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _RIO_MPORT_CDEV_H_
+#define _RIO_MPORT_CDEV_H_
+
+#ifndef __user
+#define __user
+#endif
+
+struct rio_mport_maint_io {
+       uint32_t rioid;         /* destID of remote device */
+       uint32_t hopcount;      /* hopcount to remote device */
+       uint32_t offset;        /* offset in register space */
+       size_t length;          /* length in bytes */
+       void __user *buffer;    /* data buffer */
+};
+
+/*
+ * Definitions for RapidIO data transfers:
+ * - memory mapped (MAPPED)
+ * - packet generation from memory (TRANSFER)
+ */
+#define RIO_TRANSFER_MODE_MAPPED       (1 << 0)
+#define RIO_TRANSFER_MODE_TRANSFER     (1 << 1)
+#define RIO_CAP_DBL_SEND               (1 << 2)
+#define RIO_CAP_DBL_RECV               (1 << 3)
+#define RIO_CAP_PW_SEND                        (1 << 4)
+#define RIO_CAP_PW_RECV                        (1 << 5)
+#define RIO_CAP_MAP_OUTB               (1 << 6)
+#define RIO_CAP_MAP_INB                        (1 << 7)
+
+struct rio_mport_properties {
+       uint16_t hdid;
+       uint8_t id;                     /* Physical port ID */
+       uint8_t  index;
+       uint32_t flags;
+       uint32_t sys_size;              /* Default addressing size */
+       uint8_t  port_ok;
+       uint8_t  link_speed;
+       uint8_t  link_width;
+       uint32_t dma_max_sge;
+       uint32_t dma_max_size;
+       uint32_t dma_align;
+       uint32_t transfer_mode;         /* Default transfer mode */
+       uint32_t cap_sys_size;          /* Capable system sizes */
+       uint32_t cap_addr_size;         /* Capable addressing sizes */
+       uint32_t cap_transfer_mode;     /* Capable transfer modes */
+       uint32_t cap_mport;             /* Mport capabilities */
+};
+
+/*
+ * Definitions for RapidIO events;
+ * - incoming port-writes
+ * - incoming doorbells
+ */
+#define RIO_DOORBELL   (1 << 0)
+#define RIO_PORTWRITE  (1 << 1)
+
+struct rio_doorbell {
+       uint32_t rioid;
+       uint16_t payload;
+};
+
+struct rio_doorbell_filter {
+       uint32_t rioid;                 /* 0xffffffff to match all ids */
+       uint16_t low;
+       uint16_t high;
+};
+
+
+struct rio_portwrite {
+       uint32_t payload[16];
+};
+
+struct rio_pw_filter {
+       uint32_t mask;
+       uint32_t low;
+       uint32_t high;
+};
+
+/* RapidIO base address for inbound requests set to value defined below
+ * indicates that no specific RIO-to-local address translation is requested
+ * and driver should use direct (one-to-one) address mapping.
+*/
+#define RIO_MAP_ANY_ADDR       (uint64_t)(~((uint64_t) 0))
+
+struct rio_mmap {
+       uint32_t rioid;
+       uint64_t rio_addr;
+       uint64_t length;
+       uint64_t handle;
+       void *address;
+};
+
+struct rio_dma_mem {
+       uint64_t length;                /* length of DMA memory */
+       uint64_t dma_handle;            /* handle associated with this memory */
+       void *buffer;                   /* pointer to this memory */
+};
+
+
+struct rio_event {
+       unsigned int header;    /* event type RIO_DOORBELL or RIO_PORTWRITE */
+       union {
+               struct rio_doorbell doorbell;   /* header for RIO_DOORBELL */
+               struct rio_portwrite portwrite; /* header for RIO_PORTWRITE */
+       } u;
+};
+
+enum rio_transfer_sync {
+       RIO_TRANSFER_SYNC,      /* synchronous transfer */
+       RIO_TRANSFER_ASYNC,     /* asynchronous transfer */
+       RIO_TRANSFER_FAF,       /* fire-and-forget transfer */
+};
+
+enum rio_transfer_dir {
+       RIO_TRANSFER_DIR_READ,  /* Read operation */
+       RIO_TRANSFER_DIR_WRITE, /* Write operation */
+};
+
+/*
+ * RapidIO data exchange transactions are lists of individual transfers. Each
+ * transfer exchanges data between two RapidIO devices by remote direct memory
+ * access and has its own completion code.
+ *
+ * The RapidIO specification defines four types of data exchange requests:
+ * NREAD, NWRITE, SWRITE and NWRITE_R. The RapidIO DMA channel interface allows
+ * to specify the required type of write operation or combination of them when
+ * only the last data packet requires response.
+ *
+ * NREAD:    read up to 256 bytes from remote device memory into local memory
+ * NWRITE:   write up to 256 bytes from local memory to remote device memory
+ *           without confirmation
+ * SWRITE:   as NWRITE, but all addresses and payloads must be 64-bit aligned
+ * NWRITE_R: as NWRITE, but expect acknowledgment from remote device.
+ *
+ * The default exchange is chosen from NREAD and any of the WRITE modes as the
+ * driver sees fit. For write requests the user can explicitly choose between
+ * any of the write modes for each transaction.
+ */
+enum rio_exchange {
+       RIO_EXCHANGE_DEFAULT,   /* Default method */
+       RIO_EXCHANGE_NWRITE,    /* All packets using NWRITE */
+       RIO_EXCHANGE_SWRITE,    /* All packets using SWRITE */
+       RIO_EXCHANGE_NWRITE_R,  /* Last packet NWRITE_R, others NWRITE */
+       RIO_EXCHANGE_SWRITE_R,  /* Last packet NWRITE_R, others SWRITE */
+       RIO_EXCHANGE_NWRITE_R_ALL, /* All packets using NWRITE_R */
+};
+
+struct rio_transfer_io {
+       uint32_t rioid;                 /* Target destID */
+       uint64_t rio_addr;              /* Address in target's RIO mem space */
+       enum rio_exchange method;       /* Data exchange method */
+       void __user *loc_addr;
+       uint64_t handle;
+       uint64_t offset;                /* Offset in buffer */
+       uint64_t length;                /* Length in bytes */
+       uint32_t completion_code;       /* Completion code for this transfer */
+};
+
+struct rio_transaction {
+       uint32_t transfer_mode;         /* Data transfer mode */
+       enum rio_transfer_sync sync;    /* Synchronization method */
+       enum rio_transfer_dir dir;      /* Transfer direction */
+       size_t count;                   /* Number of transfers */
+       struct rio_transfer_io __user *block;   /* Array of <count> transfers */
+};
+
+struct rio_async_tx_wait {
+       uint32_t token;         /* DMA transaction ID token */
+       uint32_t timeout;       /* Wait timeout in msec, if 0 use default TO */
+};
+
+#define RIO_MAX_DEVNAME_SZ     20
+
+struct rio_rdev_info {
+       uint32_t destid;
+       uint8_t hopcount;
+       uint32_t comptag;
+       char name[RIO_MAX_DEVNAME_SZ + 1];
+};
+
+/* Driver IOCTL codes */
+#define RIO_MPORT_DRV_MAGIC           'm'
+
+#define RIO_MPORT_MAINT_HDID_SET       \
+       _IOW(RIO_MPORT_DRV_MAGIC, 1, uint16_t)
+#define RIO_MPORT_MAINT_COMPTAG_SET    \
+       _IOW(RIO_MPORT_DRV_MAGIC, 2, uint32_t)
+#define RIO_MPORT_MAINT_PORT_IDX_GET   \
+       _IOR(RIO_MPORT_DRV_MAGIC, 3, uint32_t)
+#define RIO_MPORT_GET_PROPERTIES \
+       _IOR(RIO_MPORT_DRV_MAGIC, 4, struct rio_mport_properties)
+#define RIO_MPORT_MAINT_READ_LOCAL \
+       _IOR(RIO_MPORT_DRV_MAGIC, 5, struct rio_mport_maint_io)
+#define RIO_MPORT_MAINT_WRITE_LOCAL \
+       _IOW(RIO_MPORT_DRV_MAGIC, 6, struct rio_mport_maint_io)
+#define RIO_MPORT_MAINT_READ_REMOTE \
+       _IOR(RIO_MPORT_DRV_MAGIC, 7, struct rio_mport_maint_io)
+#define RIO_MPORT_MAINT_WRITE_REMOTE \
+       _IOW(RIO_MPORT_DRV_MAGIC, 8, struct rio_mport_maint_io)
+#define RIO_ENABLE_DOORBELL_RANGE      \
+       _IOW(RIO_MPORT_DRV_MAGIC, 9, struct rio_doorbell_filter)
+#define RIO_DISABLE_DOORBELL_RANGE     \
+       _IOW(RIO_MPORT_DRV_MAGIC, 10, struct rio_doorbell_filter)
+#define RIO_ENABLE_PORTWRITE_RANGE     \
+       _IOW(RIO_MPORT_DRV_MAGIC, 11, struct rio_pw_filter)
+#define RIO_DISABLE_PORTWRITE_RANGE    \
+       _IOW(RIO_MPORT_DRV_MAGIC, 12, struct rio_pw_filter)
+#define RIO_SET_EVENT_MASK             \
+       _IOW(RIO_MPORT_DRV_MAGIC, 13, unsigned int)
+#define RIO_GET_EVENT_MASK             \
+       _IOR(RIO_MPORT_DRV_MAGIC, 14, unsigned int)
+#define RIO_MAP_OUTBOUND \
+       _IOWR(RIO_MPORT_DRV_MAGIC, 15, struct rio_mmap)
+#define RIO_UNMAP_OUTBOUND \
+       _IOW(RIO_MPORT_DRV_MAGIC, 16, struct rio_mmap)
+#define RIO_MAP_INBOUND \
+       _IOWR(RIO_MPORT_DRV_MAGIC, 17, struct rio_mmap)
+#define RIO_UNMAP_INBOUND \
+       _IOW(RIO_MPORT_DRV_MAGIC, 18, uint64_t)
+#define RIO_ALLOC_DMA \
+       _IOWR(RIO_MPORT_DRV_MAGIC, 19, struct rio_dma_mem)
+#define RIO_FREE_DMA \
+       _IOW(RIO_MPORT_DRV_MAGIC, 20, uint64_t)
+#define RIO_TRANSFER \
+       _IOWR(RIO_MPORT_DRV_MAGIC, 21, struct rio_transaction)
+#define RIO_WAIT_FOR_ASYNC \
+       _IOW(RIO_MPORT_DRV_MAGIC, 22, struct rio_async_tx_wait)
+#define RIO_DEV_ADD \
+       _IOW(RIO_MPORT_DRV_MAGIC, 23, struct rio_rdev_info)
+#define RIO_DEV_DEL \
+       _IOW(RIO_MPORT_DRV_MAGIC, 24, struct rio_rdev_info)
+
+#endif /* _RIO_MPORT_CDEV_H_ */
index 0495884..b71fd0b 100644 (file)
@@ -354,6 +354,7 @@ header-y += reiserfs_fs.h
 header-y += reiserfs_xattr.h
 header-y += resource.h
 header-y += rfkill.h
+header-y += rio_mport_cdev.h
 header-y += romfs_fs.h
 header-y += rose.h
 header-y += route.h